Hydraulic Actuator

HydraulicActuator is a mechanical transduction device that converts the energy stored in pressurised hydraulic fluid into controlled mechanical work—linear force and stroke via hydraulic cylinders, continuous rotational torque and speed via hydraulic motors, or limited angular displacement via ro…

Semantic Classification

Content

Compositional Relationships (Components)

SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:HydraulicCylinder))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:ServoValve))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:HydraulicPump))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:Accumulator))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:PressureReliefValve))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:PositionSensor))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:ForceSensor))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:HydraulicReservoir))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:hasPart rb:HydraulicManifold))

## Dependency Relationships
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:requires rb:HydraulicPowerUnit))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:requires rb:PressurizedFluidSupply))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:requires rb:ServoController))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:requires rb:FeedbackSensor))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:requires rb:HydraulicFluid))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:requires rb:ReturnFilter))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:requires rb:HeatExchanger))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:dependsOn rb:FluidMechanics))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:dependsOn rb:ControlTheory))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:dependsOn rb:Tribology))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:dependsOn rb:BulkModulus))

## Capability Relationships
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:LeggedRobotLocomotion))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:HeavyManipulation))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:FlightSimulation))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:ExoskeletonActuation))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:CompliantForceControl))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:HighForceIndustrialAutomation))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:DeepSeaRobotics))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:enables rb:Teleoperation))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:supports rb:BostonDynamicsAtlas))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:supports rb:IITHyQFamily))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:supports rb:SarcosGuardianXO))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:supports rb:DARPARoboticsChallenge))

## Implementation Relationships
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:implements rb:PascalLawActuation))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:implements rb:ClosedLoopPositionControl))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:implements rb:ForceImpedanceControl))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:implements rb:SeriesElasticActuation))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:implements rb:ElectrohydrostaticDrive))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:implements rb:ProportionalFlowControl))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:uses rb:PIDControl))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:uses rb:ModelPredictiveControl))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:uses rb:ImpedanceControl))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:uses rb:LinearVariableDifferentialTransformer))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:uses rb:DifferentialPressureSensing))

## Reduction Relationships
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:reduces rb:TransmissionMassRequirement))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:reduces rb:GearReductionRequirement))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:reduces rb:ImpactShockLoad))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:reduces rb:ControlBandwidthLimit))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:contrastsWith rb:ElectricBLDCActuator))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:contrastsWith rb:PneumaticActuator))
SubClassOf(rb:HydraulicActuator
  ObjectSomeValuesFrom(rb:contrastsWith rb:QuasiDirectDriveMotor))

## Data Properties (Characteristics)
DataPropertyAssertion(rb:hasIdentifier rb:HydraulicActuator "RB-9011"^^xsd:string)
DataPropertyAssertion(rb:authorityScore rb:HydraulicActuator "0.87"^^xsd:decimal)
DataPropertyAssertion(rb:typicalOperatingPressureMPa rb:HydraulicActuator "21"^^xsd:decimal)
DataPropertyAssertion(rb:powerDensityKWperKg rb:HydraulicActuator "5"^^xsd:decimal)
DataPropertyAssertion(rb:forceDensityNcm2 rb:HydraulicActuator "100"^^xsd:decimal)
DataPropertyAssertion(rb:bandwidthHz rb:HydraulicActuator "30"^^xsd:decimal)
DataPropertyAssertion(rb:servoValveBandwidthHz rb:HydraulicActuator "150"^^xsd:decimal)
DataPropertyAssertion(rb:hydraulicAtlasRetirementYear rb:HydraulicActuator "2024"^^xsd:integer)
DataPropertyAssertion(rb:bulkModulusMPa rb:HydraulicActuator "1600"^^xsd:decimal)

## Property Constraints
SubClassOf(rb:HydraulicActuator
  DataMinCardinality(1 rb:operatingPressure xsd:decimal))
SubClassOf(rb:HydraulicActuator
  DataSomeValuesFrom(rb:workingFluid xsd:string))
SubClassOf(rb:HydraulicActuator
  DataAllValuesFrom(rb:requiresHPU xsd:boolean))
SubClassOf(rb:HydraulicActuator
  DataSomeValuesFrom(rb:actuatorType xsd:string))

## Annotations
AnnotationAssertion(rdfs:label rb:HydraulicActuator "Hydraulic Actuator"@en)
AnnotationAssertion(rdfs:comment rb:HydraulicActuator "Mechanical transduction device converting pressurised hydraulic fluid energy (14–35 MPa, 50–200 N/cm² force density, 2–10 kW/kg power density) into controlled motion, metered by electrohydraulic servo valves at 100–200 Hz bandwidth, deployed in legged robots (Boston Dynamics BigDog/Atlas hydraulic 2013–2024, IIT HyQ/HyQReal), exoskeletons (Sarcos Guardian XO), flight simulation platforms, deep-sea ROVs, and nuclear remote handling; DARPA Robotics Challenge 2012–2015 established hydraulic humanoids as state-of-art; Boston Dynamics Atlas retirement April 2024 and rise of electric humanoids (Tesla Optimus, Electric Atlas 2024) marks transition epoch; electrohydrostatic actuators (EHAs) converge hydraulic force density with electric cleanliness; UK research at Bath PTMC, Manchester MACE, Imperial Hamlyn Centre; hydraulics remain dominant above 5 kW/joint and 50 kg payload in 2026."@en)
AnnotationAssertion(dcterms:identifier rb:HydraulicActuator "RB-9011"^^xsd:string)
AnnotationAssertion(dcterms:subject rb:HydraulicActuator "Robotics, Fluid Power, Actuation, Legged Locomotion, Force Control, Electrohydrostatic, Servo Valve"@en)

)

Property Characteristics

AsymmetricObjectProperty(rb:requires) AsymmetricObjectProperty(rb:enables) AsymmetricObjectProperty(rb:implements) AsymmetricObjectProperty(rb:contrastsWith) TransitiveObjectProperty(rb:dependsOn) FunctionalDataProperty(rb:operatingPressure) FunctionalDataProperty(rb:powerDensityKWperKg) FunctionalDataProperty(rb:bulkModulusMPa)

About Hydraulic Actuators

• Hydraulic actuators are the enabling technology for robotic systems requiring exceptional force density, natural passive compliance under impact, and sustained high-power output at joint level. By converting pressurised fluid energy into mechanical work through Pascal’s law—force proportional to the product of gauge pressure and bore area—hydraulic actuation delivers output forces in the kilonewton range from devices weighing kilograms, without gear trains, without magnetic saturation limits, and without commutation torque ripple that characterises electric motors. This combination of high specific force, intrinsic backdrivability through fluid compressibility, and graceful overload protection through pressure relief valves has made hydraulics the foundational actuation technology for every major legged robot programme of the 2000s–2020s decade, for all commercial flight motion platforms, and for subsea and nuclear manipulators where alternative technologies are either physically inadequate or structurally impractical.
• The hydraulic actuator field has undergone a profound transformation between 2012 and 2026. The DARPA Robotics Challenge (2012–2015) placed hydraulically actuated Atlas and SCHAFT humanoids at the forefront of manipulation and locomotion demonstrations, with eleven of the top-fifteen performers relying on fluid power. By 2024, the Boston Dynamics retirement of hydraulic Atlas and simultaneous emergence of capable all-electric humanoids (Tesla Optimus Gen 2, Figure-02, Agility Robotics Digit, Unitree G1, Boston Dynamics Electric Atlas) marked the moment when electric motor technology in the 50–100 kg humanoid class matched hydraulics for locomotion—though not yet for heavy manipulation, extreme-environment deployment, or payloads above 100 kg. The electrohydrostatic actuator (EHA) architecture, borrowed from aerospace flight control, offers a convergent path: hydraulic force density without centralised HPU, fluid leakage contained within joint-scale enclosures, and efficiency competitive with geared electric drives.
• Understanding hydraulic actuators requires tracing the technology from its physical foundations through the principal subsystems, comparing it quantitatively to electric and pneumatic alternatives, and appreciating both the canonical robot platforms that defined its capabilities and the contemporary industrial niche it occupies as electric actuators claim the humanoid space.

Physical Foundations and Governing Equations

Hydraulic actuation is governed by four coupled physical principles operating simultaneously within a closed-loop control architecture.

Pascal’s Law and Force Generation

For a hydraulic cylinder with cap-end bore area A_1 and rod-end annular area A_2 = A_1 − A_rod², supply pressure P_s, return pressure P_r ≈ 0.3–1.5 MPa (back-pressure), and differential pressure ΔP = P_s − P_r:

Extension force: F_ext = P_s × A_1 − P_r × A_2 Retraction force: F_ret = P_s × A_2 − P_r × A_1

At 21 MPa supply and 1 MPa return with 50 mm bore (A_1 = 19.6 cm²) and 35 mm rod (A_2 = 9.9 cm²): F_ext = 21 × 10⁶ × 19.6 × 10⁻⁴ − 1 × 10⁶ × 9.9 × 10⁻⁴ = 41.2 − 1.0 = 40.2 kN F_ret = 21 × 10⁶ × 9.9 × 10⁻⁴ − 1 × 10⁶ × 19.6 × 10⁻⁴ = 20.8 − 2.0 = 18.8 kN

This force asymmetry between extension and retraction requires compensation in symmetric force-control algorithms; a common mitigation uses a differential (regenerative) circuit that routes rod-end return flow back to cap-end supply for extension, effectively doubling extension speed at the cost of halved force.

Orifice Flow Equation

The servo valve modulates actuator velocity by controlling volumetric flow: Q = C_d × A_v(u) × √(2ΔP/ρ) where C_d ≈ 0.64–0.70 (empirical discharge coefficient, unitless), A_v(u) is the metering edge area as a function of valve command signal u (typically ±10 V or ±10 mA), ΔP is the pressure drop across the valve land, and ρ is fluid density (855–870 kg/m³ for ISO VG 46 mineral oil at 40–60°C). Maximum spool-to-spool area A_v_max for a 40 L/min servo valve at 7 MPa drop is approximately 1.7 mm². The square-root pressure dependency means flow gain (∂Q/∂ΔP) halves as load pressure rises from zero to supply pressure—a significant nonlinearity that model-based feedforward controllers must compensate.

Hydraulic Natural Frequency and Compliance

The closed-loop hydraulic actuator behaves as a second-order system. The hydraulic natural frequency is: ω_h = √(4 β_e A²) / √(V_t m) where β_e is the effective bulk modulus of the oil-cylinder-hose system (approximately 800–1,200 MPa in practice, lower than pure oil bulk modulus of 1,400–1,700 MPa due to entrained air and hose compliance), A is cylinder bore area, V_t is total compressed oil volume (half-stroke position is worst case), and m is the effective moving mass. For HyQ leg parameters: β_e = 900 MPa, A = 15.9 cm² (45 mm bore), V_t = 50 cm³, m = 8 kg: ω_h = √(4 × 900 × 10⁶ × (15.9 × 10⁻⁴)²) / √(50 × 10⁻⁶ × 8) = √(9,100) / √(4 × 10⁻⁴) ≈ 95.4 / 0.02 ≈ 4,770 rad/s, corresponding to 760 Hz unloaded natural frequency—substantially above the servo valve bandwidth, confirming that valve bandwidth (100–200 Hz) is the effective closed-loop bandwidth limit.

Thermal Equilibrium and Efficiency

Throttling losses at the proportional valve are the dominant heat source. Hydraulic power input P_in = Q × ΔP_supply. At part-stroke (valve 50% open), 40–60% of supply pressure is dropped across the valve metering edges, with this energy converted to heat in the fluid. A typical mobile robot HPU dissipating 2.5 kW of hydraulic input power at 50% mean valve opening generates 1.0–1.5 kW of heat requiring a forced-air or water-cooled heat exchanger with 2–5 L/min coolant flow. Electrohydrostatic actuators eliminate throttling losses by matching pump displacement to demand, achieving system efficiency of 65–85% versus 40–70% for centrally throttled valve-controlled systems.

Components and Architecture

A complete hydraulic actuation system comprises five functional subsystem layers that interact under real-time closed-loop digital control.

Hydraulic Power Unit (HPU)

The HPU is the energy source for all actuator branches. In stationary industrial robots, a 3-phase electric motor (1.5–22 kW, 1,450 or 2,900 RPM) drives a bent-axis or axial-piston pump (Bosch Rexroth A10V, Parker PV series) at fixed or variable displacement. In mobile legged robots, either a petrol/diesel engine (BigDog: 15 kW 2-stroke) or a large lithium-ion battery bank driving an electric motor (Atlas hydraulic: 3.5 kW continuous / 11 kW peak from 1.5 kWh NiMH battery) serves as the prime mover. A variable-displacement load-sensing pump reduces no-load power consumption by 40–60% by automatically reducing displacement when actuator demand falls. The reservoir (5–200 L) provides fluid storage and thermal mass; the PRV (set at 105–110% of nominal operating pressure) prevents over-pressure and protects seals and hoses; the return filter (3–25 μm β₁₀ ≥ 200 absolute rating) maintains cleanliness level ISO 4406 Class 16/14/11 required for servo valve spool clearances of 1–5 μm.

Servo and Proportional Valves

Servo valves (EHSVs) are precision four-way, three-position flow-control devices constructed to tolerances of 1–5 μm spool land clearance. A two-stage torque-motor / flapper-nozzle pilot stage amplifies the low-power electrical command (±10 mA) to sufficient hydraulic force to drive the main spool against spring centering and flow forces. Moog Series 30 (30–100 L/min rated, ±20 mA, 200 Hz bandwidth), Parker D*FH, and Rexroth 4WS servo valves are standard in legged robot applications. Jet-pipe variants (Moog D792, Parker D1FP) replace the flapper-nozzle pilot with a deflecting hydraulic jet for improved contamination tolerance, particularly relevant in field-deployed robots where ISO 4406 Class 16 filtration is difficult to maintain continuously. Proportional directional control valves (PDCVs) sacrifice bandwidth (30–80 Hz) and spool precision for significantly lower cost (£300–£800 vs. £3,000–£12,000 for servo valves) and relaxed filtration requirements (ISO 4406 Class 18/16/13), suitable for industrial manipulators with positioning accuracy requirements above 0.5 mm.

Hydraulic Cylinders and Rotary Motors

Double-acting cylinders (bore 20–200 mm, stroke 25–1,000 mm) are the standard linear actuator in legged robots and industrial manipulators. Rod diameter is typically 0.6–0.8 × bore for slenderness and buckling strength. Magnetostrictive position sensors (Balluff BTL, Temposonic series) embedded in the cylinder barrel provide 1–10 μm absolute position resolution at 1–5 kHz output rate without external cable-pull transducers. Integrated pressure transducers at cap-end and rod-end ports (Kistler 4603, Hydac HDA) provide differential pressure measurements for force estimation. Axial-piston hydraulic motors (Rexroth A2FM, Parker F12, Kawasaki M series) deliver continuous rotary torque T = D_m × ΔP / (2π × η_mech), where D_m is geometric displacement (cm³/rev, 2.5–250 cm³/rev range) and η_mech ≈ 0.93–0.97. At 21 MPa and 18 cm³/rev nominal, continuous torque output is 60 Nm at η = 0.95, with peak speed 4,500 RPM and peak power 28 kW—a 7 kW/kg power density at 4 kg motor mass. Vane motors offer lower cost and quieter operation at 7–14 MPa for applications where noise is critical.

Accumulators and Energy Storage

Bladder and piston-type nitrogen-precharged accumulators (Hydac SB/SK series, Parker A-Lok) store hydraulic energy for demand transients, reducing peak pump power requirements and enabling energy recuperation. In Boston Dynamics Atlas, the accumulator bank absorbed knee-flexion energy during landing (approximately 100 J per step at 2.5 m/s running) and returned it during push-off, reducing average HPU power demand by 25–40% during dynamic gaits. The accumulator precharge pressure should be 60–70% of minimum working pressure for optimal charge/discharge energy capacity. Sizing follows ISO 6945: for a demand transient ΔV (L) between pressures P_1 and P_2 with precharge P_0, the required accumulator volume V_0 = ΔV × P_1^(1/n) / (P_2^(1/n) − P_1^(1/n)) × P_0^(−1/n), where n = 1.4 (adiabatic) for high-cycle-rate applications and n = 1.0 (isothermal) for slow-cycle accumulators.

Closed-Loop Control Architecture

A typical three-loop cascade control structure governs hydraulic actuator position control. The outer position loop (200–500 Hz sampling) compares desired position x_d(t) from the trajectory planner against measured position x(t) from the magnetostrictive sensor, generating a velocity demand v_d = K_p × (x_d − x) + K_d × (ẋ_d − ẋ) + feedforward term v_ff. The inner velocity loop (1–5 kHz) converts velocity demand to valve command signal u(t) = K_v × (v_d − v) with feedforward compensation for the nonlinear orifice flow equation u_ff = (v_d × A_piston) / (C_d × A_v_max × √(2(P_s − P_L)/ρ)), where P_L = F_load / A_piston is the load-induced pressure. Force control replaces the position outer loop with a force outer loop using differential pressure measurement: F_meas = (P_cap × A_cap − P_rod × A_rod) with a force PID or impedance controller Z_d(s) = F_d(s) / v_d(s) shaping apparent stiffness and damping. Model-based feedforward compensation for hydraulic bulk-modulus compressibility, Coulomb friction at seals (0.5–5% of rated force), and viscous drag improves tracking bandwidth from 10–20 Hz (pure PID) to 30–60 Hz (model-based plus PID). Recent deep reinforcement learning controllers (deployed on HyQReal successor in 2024 evaluation) train whole-body locomotion policies directly on hardware, bypassing explicit hydraulic models and achieving competitive rough-terrain performance—at the cost of reduced interpretability and longer initial training time.

Major Platform Families and Use Cases

Boston Dynamics BigDog and LS3 (2005–2015)

BigDog (DARPA-funded, 2005) was the first large-scale demonstration of hydraulically actuated rough-terrain quadrupedal locomotion. Key specifications: 109 kg body mass, 0.91 m tall, powered by a 15 kW single-cylinder petrol engine driving a centralised HPU at 3,000 RPM, 16 hydraulic leg actuators (4 per leg: hip ab/adduction, hip flexion, knee flexion, ankle), operating at 14 MPa. BigDog walked at 6.4 km/h on 35° slopes, ice, rubble, and deep snow—terrain where wheeled robots required human assistance. The Legged Squad Support System (LS3), derived from BigDog, carried 182 kg payload on 32 km marches in DARPA field evaluation (2012–2015), demonstrating hydraulic force advantage in load-carrying field robotics. The WildCat variant (2013) reached 32 km/h trot on flat terrain using the same hydraulic architecture. These platforms proved that hydraulic actuation could scale power across terrain obstacles that required forces no electric system of the era could sustain.

Boston Dynamics Atlas, Hydraulic Generation (2013–2024)

Atlas v1–v4 hydraulic was a 1.5 m, 75–80 kg bipedal humanoid with 28 hydraulically actuated DOF (12 legs, 7 each arm, 2 torso). The HPU operated at 35 MPa with two electric-motor-driven pump units (total 3.5 kW continuous, 11 kW peak from an onboard 1.5 kWh NiMH battery) and an accumulator bank of 2.5 L total volume. Joint actuators were custom rod-end cylinders with integrated servo valves (Moog D631 jet-pipe type, 100 L/min rated) and magnetostrictive sensors. Atlas ran at 2.5 m/s, performed backflips with 900 J peak joint power outputs impossible for electric actuators of the era, climbed obstacles 0.9 m high, and completed DARPA Robotics Challenge tasks (2015: valve operation, rubble clearing, stair climbing, vehicle driving). Boston Dynamics published over 100 refereed publications using Atlas data. On 16 April 2024, the company announced end-of-life for the hydraulic platform, posting a commemorative video (4.5 million YouTube views) alongside announcement of the electric Atlas successor featuring integrated joint actuators, Kevlar-composite links, silicon-carbide inverters, and NdFeB motor packs achieving 2–3 kW/kg at joint level compared to the hydraulic actuator’s 4–6 kW/kg—the gap having narrowed to the point where the maintenance and deployment overhead of hydraulics was no longer justified for humanoid locomotion in structured environments.

IIT HyQ Family (2010–present)

The Istituto Italiano di Tecnologia (IIT) Dynamic Legged Systems (DLS) lab in Genoa, Italy, developed the HyQ quadruped family as the primary academic platform for hydraulic legged robotics research. HyQ (2010): 70 kg, 200 bar operating pressure, 12 proportional valve-controlled hydraulic cylinders (Parker D1FP servo valves), 4 kW electric HPU; demonstrated dynamic trotting at 2.1 m/s and reactive foot placement on uneven terrain. HyQ2Max (2016): 85 kg, wider torso, improved valve bandwidth (60 Hz inner loop); published ISO 4413-compliant hydraulic circuit design guidelines. HyQReal (2019): 130 kg, titanium reinforced structure rated for 3 tonne aeroplane tow demonstrated at Genoa Caselle airport—the highest-publicised static force demonstration of any quadruped platform—requiring all four hip abduction actuators simultaneously at 28 MPa, near-rated pressure. Semini et al. published the HyQReal design in IEEE Robotics & Automation Letters (2019, 4(4), 4552–4559) documenting 175 kg tow-pull force per leg and the complete hydraulic circuit architecture. Current IIT successor (under development 2024–2026) targets EHA architecture with joint-embedded micro-pumps, retaining hydraulic force density while eliminating centralised hose routing.

Sarcos Guardian XO (2020–present)

The Sarcos Robotics Guardian XO is a 110 kg full-body powered exoskeleton using 24 hydraulic DOF (arms, torso, and legs) at 21 MPa, rated for 90 kg sustained payload lifting (4× amplification of operator effort), with 8-hour run time from an integrated lithium-ion battery pack driving an electric HPU. The Guardian XO entered commercial deployment with the US Army for logistics heavy-material handling (evaluation 2020–2021), Delta Air Lines for hangar ground support (2022–2023), and multiple energy sector partners (2024). Guardian XO competes against electric exoskeletons (Ekso Bionics EksoVest, SuitX MAX, OTTO Bock Exo) limited to 20–30 kg payload—the 3–4× force advantage of the hydraulic design is directly attributable to hydraulic force density. A guardian XO Mark III product revision announced 2024 incorporates regenerative hydraulic braking recovering 15–20% of lowering energy.

Electrohydrostatic Actuators (EHAs): Convergent Technology

EHAs embed a fixed-displacement bidirectional hydraulic pump driven by a permanent-magnet servo motor directly at the joint, drawing power from a shared DC bus (48–400 V) and controlling output force and velocity through motor speed modulation rather than valve throttling. This architecture eliminates centralised hose routing, reduces total fluid volume per DOF to 5–50 mL (eliminating large-scale leakage events), and improves round-trip efficiency to 65–85% versus 40–70% for valve-controlled centralised systems. Moog’s Compact EHA (CEHA) and Parker EHIA product lines, originally developed for aerospace primary flight control surfaces (Airbus A380 backup flight control, Boeing 787 spoilers), are being adapted for heavy robotics in 2024–2026. Joint-level power densities of 3–8 kW/kg at the shaft are achievable with modern silicon-carbide-inverter-driven PMSM pumps, competitive with valve-controlled hydraulics and 2–4× above geared BLDC actuators in the 1–10 kW range. EHAs represent the most likely replacement architecture for centralised hydraulics in new-design heavy robots between 2026 and 2032.

Flight Simulation and Haptic Interfaces

Six-DOF Stewart–Gough motion platforms for commercial flight and vehicle driving simulators universally use hydraulic servo actuation. A full-flight simulator (FFS) platform for a widebody airliner (FSTD Level D, FAA AC 120-40C / EASA CS-FSTD(A)) uses six hydraulic legs, each 1.2–1.8 m stroke, 40–80 kN peak force, at 21–28 MPa, servo-valve-controlled to ≤2 mm RMS position error across 0.01–30 Hz motion bandwidth. The Moog Motion Base MB-E-6-DOF/24 provides 24 kN per leg; CAE and L3Harris platforms (used by every major airline training programme) rely on hydraulic actuation because no electric linear actuator of comparable specification exists at competitive capital cost as of 2026. Force-feedback haptic interfaces for surgical teleoperation (Haptic Master by Moog, Phantom Desktop by 3D Systems) use miniaturised hydraulic or pneumatic actuation to render tissue compliance at 1,000 Hz update rates with 0.1 N force resolution.

Deep-Sea and Nuclear Remote Handling

Subsea intervention ROVs (Schilling Robotics Titan, Blueprint Subsea Lynx) use hydraulic manipulators rated for 6,000 m depth at 600 bar ambient. The working fluid pressure equals ambient plus 200 bar differential, maintained by a pressure-compensation bladder. Electric motors at this depth require thick-wall pressure housings adding 5–15 kg per joint; hydraulic cylinders are inherently pressure-equalised and require only O-ring face seals rated for their differential operating pressure. UK Atomic Energy Authority (UKAEA) and NDA programmes at Sellafield and Dounreay use hydraulic manipulators for nuclear decommissioning where electric motors cannot be shielded against 10 kGy radiation doses without heavy lead encapsulation; hydraulic cylinders and motors tolerate radiation to >100 MGy with appropriate seal material selection (PTFE and Viton degrade above 200 kGy; EPDM tolerates to 500 kGy; ceramic seals to >1 MGy).

Comparison to Electric BLDC and Pneumatic Actuators

A systematic quantitative comparison is essential for actuator selection in robotic system design.

Hydraulic vs. Electric BLDC with Gearbox

Electric BLDC motors with harmonic-drive (reduction 50:1–160:1, e.g. Harmonic Drive CSF/SHF series) or planetary-gearbox (10:1–100:1, e.g. Maxon, Neugart) transmissions dominate humanoid robots from 2020 onward. Advantages over hydraulics: zero fluid leakage enabling deployment in hospitals, food processing, and office environments; acoustic noise below 50 dB; no HPU mass overhead; single cable routing per joint versus hose plus wire; regenerative braking recovering 15–25% of descent energy; mature high-volume supply chains reducing per-joint cost to £300–£3,000. Disadvantages: power density at joint shaft limited to 0.5–1.5 kW/kg for harmonic-drive combinations (gear losses 15–25%, motor copper losses) versus 3–8 kW/kg hydraulic; harmonic-drive compliance and hysteresis (40–60 Nm backlash-equivalent stiffness) degrading transparent force control in delicate assembly; thermal current limits restricting sustained high-force output (motors derate above 60–80°C winding temperature, reaching temperature limits in 2–10 min of full-torque operation).

Quasi-direct-drive (QDD) motors (MIT Cheetah series, Unitree A1/Go1, ANYmal electric joints) reduce gear ratio to 6:1–10:1 using high-pole-count BLDC designs (6–36 pole pairs) with segment-wound stators. QDD achieves 0.5–2.5 kW/kg joint power density, dramatically improved backdrivability (reflected inertia 36–100× motor inertia vs. 2,500–25,000× for harmonic drives), sub-millisecond torque bandwidth from current-controlled inverters, and zero gear hysteresis. The MIT Cheetah 3 and Mini Cheetah demonstrated that electric QDD quadrupeds could match the running speed (3 m/s trot), jumping height (0.9 m), and rough-terrain performance of hydraulic counterparts in the <40 kg body mass class—the watershed result published in IEEE T-Robotics 2019 (Wensing et al.) that accelerated the electric transition for small-to-medium legged robots.

The breakeven point in power density at joint shaft between hydraulic valve-controlled and electric QDD actuators lies at approximately 3–5 kW output power per joint. Below this threshold, modern QDD motors offer competitive or superior power density with greater efficiency, zero leakage, and lower noise. Above 5–10 kW per joint—required for heavy manipulation, large exoskeletons, and construction robots—hydraulic or EHA systems retain a decisive advantage through 2030.

Hydraulic vs. Pneumatic

Pneumatic actuators operate on compressed air at 0.4–1.0 MPa (60–145 psi), delivering force densities 20–50× lower than hydraulic at equivalent bore (pneumatic 1–15 N/cm² vs. hydraulic 50–200 N/cm²). Air compressibility (bulk modulus approximately 0.1–0.2 MPa, versus 800–1,700 MPa for hydraulic oil) makes precise position control difficult without expensive proportional valves and model-based compensation for compressibility, and limits closed-loop position bandwidth to 5–25 Hz in practice. However, these same properties—high compliance, very low reflected inertia, intrinsically safe interaction forces—make pneumatics advantageous for soft robotics, rehabilitation exoskeletons requiring gentle interaction with human limbs, agricultural gripping of delicate produce, and assistive devices prioritising safety over performance. McKibben pneumatic artificial muscles (braided sleeve actuators, Festo MAS/DMSP series) deliver 40–60% contraction at 0.3 MPa with compliance closely matching biological muscle pennation mechanics; they have been used in rehabilitation robots (Shadow Air Muscles, Festo Bionic Arm) where series elastic compliance is more important than bandwidth.

Quantitative Comparison Table (2026 State of Practice)

Metric	Hydraulic (valve)	EHA	BLDC+Harmonic	BLDC+QDD	Pneumatic
Power density joint (kW/kg)	3–8	3–8	0.3–1.5	0.5–2.5	0.05–0.2
Force density (N/cm²)	50–200	50–150	8–40	5–30	1–15
Bandwidth (–3 dB) Hz	10–60	15–80	20–80	50–200	2–20
Backdrivability	High	High	Low (harmonic)	High	High
Leakage risk	High	Low	None	None	Low (oil mist)
Peak noise dB	70–90	50–70	45–65	40–60	55–75
Round-trip efficiency	40–70%	65–85%	60–80%	75–95%	15–40%
Deployment environment	Industrial/outdoor	Industrial/outdoor	Office/clean	Office/clean	Medical/agri
Payload crossover (kg)	>50 advantage	>50 advantage	<50 competitive	<40 competitive	<10 kg

Academic Context and Foundational Research

The theoretical foundation of hydraulic control systems was established by Herbert E. Merritt’s 1967 textbook Hydraulic Control Systems (Wiley), which derived the linearised transfer function of a valve-controlled hydraulic cylinder as a second-order underdamped system with hydraulic natural frequency ω_h = √(4β_e A²/V_t m) and damping ratio ζ_h = (C_e/2A)√(m/β_e V_t) + (D_m/(4A V_t β_e m)), where C_e is equivalent leakage coefficient. This framework, extended by Jelali and Kroll (2003) to nonlinear identification and model predictive control, remains the standard analytical tool for servo-hydraulic loop design.

Pratt and Williamson’s 1995 IROS paper introducing series elastic actuators (SEA) — placing a physical compliant spring in series between actuator gearbox and load, measuring force through spring deflection — was conceptually motivated by the need for safe force control; hydraulic systems achieve the same passive compliance through oil bulk modulus, without an additional spring element, and with lower reflected inertia than geared electric SEAs. The conceptual parallel between oil compressibility and spring compliance was formalised by Boaventura et al. (2012) in their IEEE T-Robotics paper on model-based hydraulic impedance control for HyQ.

The DARPA Robotics Challenge (2012–2015) was the field’s landmark evaluation. Team IHMC’s Atlas (hydraulic) won 2nd place; Kaist DRC-HUBO (electric) won 1st; Team MIT Atlas placed 3rd; SCHAFT (hydraulic) placed 2nd in DRC trials. Post-DRC analysis (Atkeson et al. 2016, Field Robotics) noted that hydraulic systems enabled robust joint torque control without torque sensors, and their passive compliance absorbed unpredictable impact loads during falls (11 teams’ robots fell during the 2015 finals). The post-DRC period (2016–2020) saw rapid maturation of electric QDD actuation—Seok et al. (2015, IEEE/ASME T-Mechatronics) demonstrated that gear ratio optimisation for MIT Cheetah could approach hydraulic performance metrics in the 100 W–1 kW range—leading to the field’s divergence into electric platforms for agile locomotion and retained hydraulics for high-payload manipulation.

Christopoulos and Tsagarakis at IIT (2006–2014) developed impedance control frameworks for hydraulic legs, while Gehring et al. at ETH Zürich extended whole-body control to hydraulic quadrupeds, subsequently migrating to electric actuation in ANYmal (first release 2016, the most commercially successful legged robot as of 2026 with 500+ units deployed globally, all-electric). The University of Edinburgh Robotics group (Vijayakumar, Kormushev) contributed learning-based whole-body controllers applied to hydraulic platforms between 2010 and 2018, developing Bayesian optimisation methods for hydraulic parameter identification that reduced calibration time from 4 hours to 25 minutes.

Current Landscape (2026)

The 2026 landscape is characterised by clear bifurcation between the humanoid/agile locomotion segment (electric BLDC dominant) and the heavy manipulation, exoskeleton, and extreme-environment segment (hydraulic and EHA dominant).

Electric humanoid consolidation: Tesla Optimus Gen 2 (2024), Boston Dynamics Electric Atlas (2024), Agility Robotics Digit v5 (2025), Figure-02 (2024), Apptronik Apollo (2024), and Unitree G1/H1 all use fully electric actuation, selling to automotive manufacturing, logistics, and semiconductor fabrication markets. Total global humanoid deployment reached approximately 10,000 units (across all makes) by end-2025, all electric. Hydraulic humanoids have zero commercial deployment as of 2026.

Heavy industrial hydraulic incumbency: Kuka KR1000 Titan (1,000 kg payload) and comparable Fanuc M-2000 series use hydraulic counterbalance cylinders and electric joints; full-hydraulic manipulators above 200 kg payload (Rexroth, Moog, Parker custom arms for nuclear and aerospace assembly) remain hydraulic or hybrid hydraulic-electric. Caterpillar, Komatsu, and Volvo hydraulic construction equipment represents the largest installed base of hydraulically actuated robotic machinery globally, totalling >2 million units.

Exoskeleton market split: Sarcos Guardian XO (hydraulic, 90 kg payload) and Hyundai VEX (hydraulic, 100 kg payload for vehicle assembly) occupy the high-payload industrial segment; Ekso, SuitX, and OTTO Bock electric exoskeletons dominate medical rehabilitation and moderate-load industrial assistance below 30 kg payload.

EHA development pipeline: Moog, Parker, and start-up Nuvera are actively developing second-generation EHA products targeting 5–10 kW/kg joint power density with additive-manufactured titanium manifolds and silicon-carbide-inverter efficiency of 97.5%. First EHA-jointed heavy robots are expected in field evaluation by 2027–2028. IIT’s HyQReal successor and a Toyota Research Institute heavy arm programme represent the leading academic EHA programmes in legged and manipulation robotics respectively.

UK Context

University of Bath Centre for Power Transmission and Motion Control (PTMC): The Bath PTMC group (Professors Andrew Plummer, Nigel Johnston, and colleagues) is Europe’s leading academic centre for digital hydraulics, variable-speed pump control, and hydraulic energy recuperation. The group has published foundational work on digital displacement technology (binary-coded valve arrays), servo-solenoid valve characterisation, and hydraulic actuator fault diagnosis. Key collaborative projects include the Aerospace Technology Institute–funded programme on hydraulic actuation for aircraft primary flight control with Moog and Airbus, and the EPSRC-funded project on energy recuperation in mobile hydraulic machinery with JCB.

University of Manchester MACE (Mechanical, Aerospace and Civil Engineering): The Manchester fluid power group researches servo-valve dynamics, digital hydraulic switching, and model-based control of hydraulic systems for industrial robots. They collaborate with Bosch Rexroth’s UK application engineering centre in Gillingham (Kent), the largest hydraulic component engineering site in the UK. The MACE group applies machine learning to hydraulic fault detection and predictive maintenance for manufacturing plant, with several published results on anomaly detection in hydraulic press circuits using LSTM classifiers trained on pressure sensor data.

Imperial College London — Hamlyn Centre for Robotic Surgery: The Hamlyn Centre operates miniaturised hydraulic and pneumatic actuators in surgical manipulation devices and flexible endoscopic instruments. Research on fluid-driven continuum robots and hydraulic catheters (Kaspar Althoefer, King’s College London; previously Imperial) has contributed novel micro-hydraulic silicon-tube actuators for single-port laparoscopic tools. The hydraulic advantage in surgical robots is electrical isolation: fluid-power actuators remote from patient eliminate electrosurgical interference, relevant in MRI-guided interventions where electrical actuation causes significant artefact.

Northern England industrial base: Rotork plc (Bath, with significant engineering in Leeds) manufactures hydraulic and electro-hydraulic valve actuators for oil, gas, power, and water infrastructure; the IQ3/CVA product range includes CANopen/PROFIBUS-networked electrohydraulic actuators deployed across North Sea platform valve control. Parker Hannifin UK (Tewkesbury and Barnstaple) produces servo valves, hydraulic cylinders, and EHA assemblies for aerospace and defence robotics. BAE Systems’ Brough (Yorkshire) and Warton (Lancashire) sites use hydraulic actuation for armoured vehicle assembly robots and evaluate hydraulic walking platforms for mine-clearance under MOD DASA contracts. The Advanced Manufacturing Research Centre (AMRC) at Sheffield applies additive titanium manufacturing to reduce hydraulic manifold mass by 30–45% in aerospace robot tooling programmes. JCB at Rocester (Staffordshire), as the UK’s largest hydraulic equipment manufacturer, indirectly drives the UK hydraulic supply chain—seal suppliers, pump manufacturers, and valve assemblers—concentrated in the West Midlands and Yorkshire.

Regulatory framework: UK HSE guidance HSG244 (Safe Use of Hydraulic Equipment, 2014) governs maintenance intervals, fluid safety data sheets, and pressure testing requirements. The Pressure Systems Safety Regulations 2000 (PSSR 2000) require a written scheme of examination for any hydraulic system with stored energy exceeding 250 bar·L, limiting robot deployment flexibility without site engineering assessment. The UK’s adoption of retained EU Machinery Directive 2006/42/EC (now UK Machinery Regulations 2008, amended) mandates hydraulic circuit risk assessment under EN ISO 13849 safety integrity level analysis, requiring hydraulic safety blocks with fail-safe de-energise for any robot joint capable of injuring an operator.

Future Directions (2026–2030)

Electrohydrostatic actuator maturation: Joint-embedded EHAs with additive-manufactured titanium manifolds, silicon-carbide inverter drives, and ester-based biodegradable working fluids are projected to achieve 7–9 kW/kg joint power density by 2029, matching best-case hydraulic valve-controlled performance while eliminating HPU centralisation. The key remaining technical barriers are micro-pump durability (targeted 10 million full-pressure cycles), seal reliability at rapid thermal cycling from cold-start to operating temperature, and acoustic noise from on-demand pump operation inside robot links. Parker and Moog second-generation EHA products planned for 2027–2028 commercial release target these metrics.

Digital hydraulics and switching valve arrays: High-speed on-off solenoid valves (1–10 ms switching, Parker Screw-In Cartridge D1VSF series) modulated by pulse-width or duty-cycle control can synthesise proportional flow without continuous throttling, achieving 90–97% transmission efficiency versus 40–70% for proportional valve throttling. Bath PTMC group has demonstrated 96% efficiency digital hydraulic drivetrain in simulation and bench testing (2023). Commercial digital displacement pumps (Danfoss Digital Displacement®, first commercial units 2022) apply the same principle to variable-displacement pumping. Full digital hydraulic actuation systems for heavy robots are expected in prototype evaluation by 2028.

Biodegradable and fire-resistant working fluids: ISO 15380 Type HETG (triglyceride-based vegetable hydraulic fluid) and HFDU (synthetic ester) fluids achieve 70–80% of mineral oil bulk modulus while meeting EN ISO 11158 biodegradability (>60% in 28 days) and EU Ecolabel requirements. Bosch Rexroth Hees Synfluid and Shell Naturelle HF-E series are commercially available; Bath PTMC published cavitation characteristics and thermal degradation rates for HETG in 2024. EU Machinery Regulation 2023/1230 effective 2027 will require justification for mineral oil use where fire risk is present, accelerating ester fluid adoption in industrial robots.

AI-augmented hydraulic dynamics learning: Deep reinforcement learning policies trained on physics-based hydraulic simulation (MuJoCo with Rexroth servo valve model, bulk-modulus contact compliance) and fine-tuned on hardware are replacing explicit fluid model identification. NVIDIA Isaac Lab (2024 release) includes a hydraulic joint plugin modelling nonlinear valve flow and compressibility; the IIT DLS group is using Isaac-trained policies for HyQReal successor locomotion trials. This approach eliminates the 2–4 hour hydraulic parameter calibration procedure for new robot deployments.

Cryogenic and space hydraulics: NASA’s 2030s Europa Lander surface arm concept uses water-glycol hydraulic fluid (freezing point suppressed to −60°C) compatible with cryogenic temperatures where petroleum oils solidify, with titanium seals tolerating radiation. ESA’s ExoMars 2030 sample arm evaluation has included hydraulic cylinder actuation as a fallback for environments where electric motor bearings seize at −120°C. This represents hydraulic actuation’s frontier application space where electric actuators face fundamental thermal and radiation constraints.

Key Technical Specifications Reference (2026 State of Practice)

Representative hydraulic actuator system specifications for robotic applications, drawn from published platform data and industry datasheets:

Boston Dynamics Atlas v4 (hydraulic, 2019–2024):

• Body mass: 80 kg; height: 1.5 m; DOF: 28 hydraulic

• Operating pressure: 350 bar (35 MPa)

• HPU: electric motor, 3.5 kW continuous / 11 kW peak

• Battery: 1.5 kWh NiMH; runtime: 60–90 min walking

• Servo valves: Moog D631 jet-pipe, 100 L/min, 120 Hz bandwidth

• Joint bandwidth: 30–40 Hz (outer position loop)

• Running speed: 2.5 m/s; backflip achieved 2018

  IIT HyQReal (2019):

• Body mass: 130 kg; DOF: 12 hydraulic (3 per leg)

• Operating pressure: 200 bar (20 MPa)

• HPU: 4 kW electric; proportional valves (Parker D1FP)

• Maximum pull force demonstrated: >3,000 kg (aeroplane tow, Genoa 2019)

• Joint force capacity: 175 kg per leg hip abduction at rated pressure

• Valve bandwidth: 60 Hz inner loop; 15 Hz outer position loop

  Sarcos Guardian XO (2021 commercial):

• Total mass: 110 kg (exoskeleton); payload: 90 kg (4× amplification)

• DOF: 24 hydraulic (full body: arms, torso, legs)

• Operating pressure: 210 bar (21 MPa)

• Power source: lithium-ion battery, 8-hour runtime

• Force response latency: <5 ms closed-loop

• Acoustic noise: 62 dB at operator position

  Moog Series 30 Servo Valve (reference component):

• Rated flow: 38 L/min at 70 bar drop (ΔP = 7 MPa)

• Bandwidth (–90°): 100 Hz at 25% rated flow; 200 Hz at 5% rated flow

• Hysteresis: <0.5% rated input

• Threshold: <0.1% rated input

• Null leakage: <0.5 L/min at 210 bar supply

• Spool clearance: 3–4 μm (requires ISO 4406 Class 16 filtration)

• Operating temperature: −54°C to +107°C (with appropriate fluid)

• Mass: 1.1 kg; price (2024): £8,500–£12,000 per valve

  Comparison: BLDC QDD Joint (MIT Mini Cheetah reference, 2021):

• Motor mass: 0.3 kg; joint mass with gearbox: 0.55 kg

• Peak torque: 17 Nm at 22 A; continuous: 3 Nm at 6 A

• Gear ratio: 6:1 planetary; reflected inertia: 0.018 kg·m²

• Bandwidth: >100 Hz torque; position accuracy: ±0.1° (encoder)

• Power density: 0.8 kW/kg at peak; continuous: 0.2 kW/kg

• Cost: £450–£900 per joint (2024 production volume)

Hydraulic Fluid Selection and System Chemistry

The working fluid is the lifeblood of the hydraulic actuation system, transmitting force, lubricating moving components, dissipating heat, and protecting metal surfaces from corrosion. Fluid selection is not merely a procurement decision but a systems-engineering choice that determines achievable performance, maintenance schedule, environmental liability, and seal material compatibility. Poor fluid selection or inadequate fluid maintenance causes an estimated 70% of hydraulic system failures in industrial plant (Bosch Rexroth field study, 2019).

The working fluid is as critical to hydraulic actuator performance as the mechanical components it pressurises. Fluid selection determines achievable operating pressure, viscosity-temperature behaviour, seal compatibility, fire resistance, and environmental risk profile.

Mineral Hydraulic Oils

Mineral oil base stocks (Group I–III solvent refined, hydrocracked, or hydrotreated paraffinic stocks) blended with antiwear (zinc dialkyldithiophosphate, ZDDP), oxidation inhibitor, rust and corrosion inhibitor, demulsifier, and viscosity index improver packages constitute 80–85% of hydraulic fluid used globally. ISO VG 46 (46 cSt at 40°C, 6.5 cSt at 100°C, viscosity index 95–105) is the most widely used grade for robotics. Advantages include low cost (£0.80–£2/L bulk), excellent lubricity extending pump and valve life, high bulk modulus (1,400–1,700 MPa at 40°C), wide availability, and compatibility with nitrile (NBR) and polyurethane seals. Disadvantages are fire risk (flash point 180–220°C for Groups I–III), environmental persistence (inherently biodegradable score 15–25% versus >60% required for EU Ecolabel), and aquatic toxicity (LL50 > 1,000 mg/L for mineral oil, technically low acute toxicity but chronic bioaccumulation concern from PAH components in Group I stocks).

Synthetic Ester and Biodegradable Fluids

ISO 15380 classifies four categories of environmentally acceptable hydraulic fluids (EAHFs): HETG (triglyceride-based, natural ester), HEES (synthetic ester, fully saturated), HEPG (polyalkylene glycol, PAG), and HEPR (polyalphaolefin/related hydrocarbon). HEES synthetic esters (e.g. Bosch Rexroth Hees Synfluid 46, Shell Naturelle HF-E 46, Mobil EAL Hydraulic 46) achieve bulk modulus 1,200–1,500 MPa (10–20% lower than mineral oil, increasing hydraulic natural frequency and compliance), viscosity index 150–180 (superior temperature stability), biodegradability >60% in 28 days (OECD 301B), low aquatic toxicity (LL50 > 10,000 mg/L), and fire point >260°C. Seal compatibility requires transition from NBR (swells in ester) to FKM (Viton) or PTFE seals, adding seal replacement cost at system conversion. EAHF adoption in outdoor-deployed hydraulic robots increased from 8% in 2018 to approximately 22% in 2025 as EU Ecolabel incentives and UK Environment Agency guidance strengthened.

Fire-Resistant Hydraulic Fluids

HF-A (oil-in-water emulsion, 95% water), HF-B (water-in-oil invert emulsion), HF-C (water-glycol, ASTM D7662 Type HFGS), and HF-D (phosphate ester, Fyrquel EHC, Skydrol) categories serve applications requiring superior fire resistance: die casting, steel mills, aircraft hydraulics, underground mining. HF-C water-glycol fluids (bulk modulus 3,000–4,000 MPa at 40°C, 2–3× higher than mineral oil) require lower system pressures (14 MPa maximum) due to pump material compatibility (aluminium components dissolve in alkaline glycol solutions) but provide excellent fire resistance and pump-down-to-stop anti-siphon performance. Phosphate ester HF-D fluids are used in aircraft primary flight control EHA systems (Airbus A380 backup hydraulics) where mineral oil fire risk near hot engine components is unacceptable; their very high bulk modulus (2,800–3,200 MPa) provides exceptionally stiff hydraulic circuits ideal for precision flight control.

Fluid Condition Monitoring

Real-time hydraulic fluid condition monitoring has become standard in high-availability robot systems. Inline sensors measure: viscosity (Cambridge Viscosity ViscoScope, Anton Paar L-Vis 510), particle count (Pall icount PDR2, 4–70 μm channels at ISO 4406), water content (capacitive sensors, 0–2,000 ppm range), and temperature (Pt100 RTD, ±0.1°C). Ferrographic oil analysis (offline laboratory, monthly intervals in high-duty cycle systems) quantifies metallic wear debris type (ferrous = pump/motor wear; aluminium = valve body wear; copper = bronze bearing wear) providing early warning of component failure. UK oil analysis laboratories (Spectro Scientific, Jorin, Oil Analysis Laboratories at Derby and Birmingham) offer 48-hour turnaround routine analysis services for industrial robot fleets. Integration of fluid condition data with digital twin models enables predictive maintenance scheduling that extends HPU service intervals by 40–60% compared to calendar-based OEM recommendations.

Risks, Limitations, and Failure Modes

Hydraulic actuation carries a well-characterised and manageable set of failure modes. Each failure mode has established detection and mitigation strategies; the risk profile is not prohibitive but does require engineering investment that electric systems avoid.

Hydraulic actuation carries a well-characterised set of failure modes and operational risks that system designers must address through redundancy, monitoring, and maintenance schedules.

Fluid Leakage and Environmental Contamination

Hydraulic fluid leakage is the most common failure mode in field-deployed hydraulic robots and the primary environmental concern. Mineral hydraulic oil (ISO VG 46) is classified as a Category 3 water pollutant in Germany (WGK 1–2) and regulated under the UK Water Resources Act 1991 and Environment Agency guidelines. A single O-ring seal failure at 21 MPa can release 0.5–5 litres of oil per minute onto soil or into waterways before operator response. Estimated annual hydraulic oil leakage from UK hydraulic equipment is 50,000–120,000 tonnes (BFPA 2021 estimate). For legged robots deployed in outdoor environments (DARPA field trials, forestry, agriculture), leakage monitoring using inline flow sensors comparing pump output to actuator consumption, and automatic HPU shutdown on leakage detection above a threshold flow imbalance, are essential. ISO 4413 Clause 5.3 mandates that hydraulic systems in mobile machinery include leakage detection and automatic isolation. Synthetic ester and vegetable-based hydraulic fluids (ISO 15380 HETG, HEES categories) reduce environmental impact while maintaining 80–90% of mineral oil performance characteristics; their adoption in outdoor hydraulic robots has increased from approximately 5% of installations in 2015 to an estimated 20% in 2025.

Contamination and Servo Valve Failure

Servo valve spool clearances of 1–5 μm are susceptible to particle contamination in hydraulic fluid. A single 15 μm particle (half the diameter of a human hair) can wedge between spool land and bore, causing valve sticking (hysteresis above 5% FS), valve oscillation, or complete spool seizure. ISO 4406 Class 16/14/11 (≤1,300 particles ≥4 μm, ≤320 ≥6 μm, ≤40 ≥14 μm per 100 mL) is the minimum cleanliness requirement for servo valves; achievement requires both offline (kidney loop) and inline (high-pressure) filtration at 3–6 μm absolute ratings. Contamination is the cause of approximately 70% of servo valve failures in industrial hydraulic systems (Rexroth field data 2019). In legged robots operating outdoors, maintaining ISO 4406 Class 16 is challenging; Moog jet-pipe and Rexroth nozzle-sleeve valve variants offer improved contamination tolerance at the cost of 20–30% reduced bandwidth compared to flapper-nozzle designs.

Thermal Management

Hydraulic system operating temperature must be maintained in the 40–60°C range for optimal oil viscosity (ISO VG 46: 46 cSt at 40°C, 6.5 cSt at 100°C). Above 80°C, mineral oil oxidation rate doubles per 10°C increase (Arrhenius relationship), accelerating additive depletion and varnish deposition on valve spools. Below 10°C, high viscosity increases pressure drop across filters, starving pumps and increasing valve response time. In stationary industrial robots, a thermostatically controlled air-blast heat exchanger maintains temperature; in mobile robots, waste heat from throttling losses is a significant design challenge. Boston Dynamics Atlas operated with a forced-air cooling system integrated into the torso, with temperature sensors triggering reduced-performance modes above 65°C HPU return-line temperature. Thermal analysis using computational fluid dynamics is now standard practice in hydraulic robot thermal design (ANSYS Fluent, MATLAB/Simscape Fluids models).

Noise and Vibration

Hydraulic servo valves generate acoustic noise from turbulent flow at orifice edges (70–90 dB at 500 mm distance) and from HPU pump ripple (typically 0.1–3% pressure ripple at pumping frequency). Pump ripple propagates through the hydraulic circuit as fluid-borne noise, exciting structural resonances in robot links and generating airborne noise. This limits hydraulic robot deployment in quiet environments (medical facilities, offices, homes). Attenuation strategies include: side-branch Helmholtz resonators tuned to pump fundamental frequency; viscoelastic hose materials with high internal damping; flexible hose loops isolating HPU vibration from the rigid manifold; and variable-displacement pumps whose smooth flow output eliminates ripple at partial displacement settings. The Sarcos Guardian XO achieves 62 dB operation (a noise level comparable to a normal conversation) through combination of pump anti-vibration mounts, encapsulated HPU, and smooth-flow variable-displacement pump—demonstrating that hydraulic noise can be engineered to acceptable levels for close human interaction.

Seal Degradation and Maintenance Schedules

Hydraulic cylinder and motor seals (polyurethane rod seals, PTFE and HNBR piston seals) have finite service lives of 500–5,000 operating hours depending on speed, pressure cycling, and fluid compatibility. In legged robots with 500–2,000 h/year operating time, seal replacement every 6–24 months is typical. IIT HyQ maintenance records (Semini 2017 field robotics appendix) document seal replacement intervals of 800–1,200 h for the knee and hip cylinders operating at 200 bar with 5 Hz pressure cycling. Condition monitoring through quantitative ferrographic oil analysis (detecting metal particles from seal and bearing wear), inline particle counters (Pall icount PDR, Hach-Lange PCME), and real-time seal leakage monitoring through return-line flow sensors enables predictive maintenance that extends between-service intervals by 30–50% compared to calendar-based replacement.

Control System Design Patterns for Hydraulic Robots

Hydraulic actuation requires specialised control architectures that account for nonlinear fluid dynamics, pressure-dependent flow gain, and the coupling between joint forces and system pressure.

Cascade Control Structure

The standard three-loop cascade for a hydraulic joint operates as follows:

Outer loop — position or force (200–500 Hz)

• Desired position x_d or force F_d from motion planner or operator command

• Measured position from magnetostrictive or optical encoder sensor

• PID error law: v_d = K_p(x_d − x) + K_i∫(x_d − x)dt + K_d(ẋ_d − ẋ) + v_ff

• Feedforward term v_ff computed from trajectory derivatives reduces steady-state lag

  Middle loop — velocity (1–5 kHz)

• Converts velocity demand v_d to valve command u(t) accounting for pressure-dependent flow gain

• Nonlinear inverse: u = (v_d × A_eff) / (C_d × A_v_max × √(2(P_s − P_L)/ρ))

• P_L = (F_ext / A_cap − P_r × A_rod/A_cap) is load-induced cylinder pressure

• Pressure P_s and P_L measured at 4 kHz by piezoelectric transducers at cylinder ports

  Inner loop — valve spool position (10–20 kHz in modern digital servo drives)

• Servo valve has internal LVDT spool position feedback at torque-motor stage

• Provides linearisation of spool–flow relationship and compensation for flow forces on spool

  Force and Impedance Control

  Hydraulic robots excel at direct force control because the working fluid provides differential pressure measurement of output force without a separate force sensor:

  F_joint = P_cap × A_cap − P_rod × A_rod − F_friction(v) − F_seal

  where F_friction(v) is velocity-dependent viscous drag (typically 1–5% of rated force) and F_seal is static seal breakout force (0.5–3% of rated force, compensated by deadband in the control law). Force control bandwidth achievable with pressure-based measurement is 30–80 Hz, limited by compressibility of the oil column rather than sensor dynamics. For comparison, strain-gauge load cells in electric joint designs achieve 1,000+ Hz force bandwidth but require calibration and are vulnerable to lateral loading and thermal drift. Impedance control shapes the apparent mechanical impedance Z(jω) = F(jω)/v(jω) of the joint to a desired target impedance Z_d = K_d/(jω) + B_d (spring-damper), enabling: soft contact mode (Z_d → low K_d, low B_d) for delicate assembly, stiff position mode (Z_d → high K_d) for trajectory following, and tuned damping for shock absorption during landing impacts.

  Model Predictive Control for Hydraulic Systems

  Model predictive control (MPC) applied to hydraulic actuators solves a finite-horizon optimal control problem at each sampling instant, optimising valve commands over a prediction horizon of T_p = 50–200 ms to minimise a cost function combining tracking error, valve control effort, and hydraulic pressure constraints:

  J = Σ [||x(k) − x_d(k)||²_Q + ||u(k)||²_R + λ_P × max(0, P(k) − P_max)²]

  where Q and R are weighting matrices for tracking and effort, and the pressure constraint term penalises commands that would drive system pressure above the PRV setpoint. MPC naturally handles actuator saturation (valve command bounds ±100%), pressure limits, and velocity limits in a unified framework. For hydraulic legged robots, Boaventura et al. (2014 IEEE T-Robotics follow-on work) demonstrated that MPC with a 100 ms prediction horizon at 200 Hz sample rate achieved 35% lower energy consumption than PID during dynamic trotting on HyQ, by preemptively reducing flow demand ahead of predicted load transitions. Computational cost on ARM Cortex-A72 hardware (Raspberry Pi 4 class) is approximately 0.4 ms per solve for a 12-DOF hydraulic quadruped, within the 5 ms budget for the 200 Hz outer loop.

  Whole-Body Control with Hydraulic Compliance

  Whole-body controllers (WBC) for hydraulic legged robots formulate a hierarchical quadratic program (HQP) that simultaneously satisfies constraints across all robot joints, using the natural compliance of hydraulic actuators as a regularisation term. The HQP minimises joint torque commands τ = A × F_contact + b subject to: equation-of-motion constraints, contact constraints (no slipping or lifting), joint torque limits, and actuator bandwidth limits. For hydraulic robots, the effective joint stiffness K_hyd = 4β_e A²/V_t from oil compressibility contributes passive compliance that acts as implicit regularisation, allowing the WBC to generate more aggressive whole-body motions without violating contact stability constraints. This property was identified as a key advantage of hydraulic Atlas over early electric humanoids in the post-DRC literature (Wensing et al. 2017 WBC comparison supplement).

Hydraulic Actuation in Humanoid Robotics: The Transition Decade (2012–2024)

The decade from 2012 to 2024 constitutes the complete arc of hydraulic humanoid robotics as a primary research and commercial technology—from the DARPA Robotics Challenge announcement through the Boston Dynamics Electric Atlas successor reveal. Understanding this transition illuminates both the genuine advantages that hydraulics provided and the specific technological developments that made electric actuation competitive.

DARPA Robotics Challenge Context (2012–2015)

The DARPA Robotics Challenge (DRC) Trials (December 2013) and Finals (June 2015) represented the field’s most demanding public evaluation of humanoid robot capabilities. The primary motivation was developing robots capable of operating in disaster environments—nuclear plant accidents, chemical spills—where terrain is unstructured and forces unpredictable. DARPA selected Atlas v1 as the shared platform provided free to 7 teams (Team IHMC, Team Draper, Team MIT, Team WPI-CMU, Team TRACLabs, Team Lockheed, Team Trooper), ensuring that 7 of the 25 DRC Teams used the same hydraulic platform. Team SCHAFT (acquired by Google in 2013, later SoftBank then released) used their own hydraulic design with an integrated HPU and achieved 2nd place in Trials. The hydraulic advantage was multifaceted: joint torque density enabled reliable door opening (typical 10–25 Nm), valve-turning (8–15 Nm at 350 mm radius), debris clearing (50–100 N pushing force), and driving (steering wheel torque 5–12 Nm at full lock)—forces that exhausted electric motor thermal limits in multi-minute sustained operation. Natural compliance absorbed unpredictable impact loads during the 23 documented falls across all teams in the 2015 Finals, preventing joint damage that would have been catastrophic in rigid-gearbox electric designs.

Team KAIST DRC-HUBO, which won the 2015 Finals with time 44 minutes and zero falls, was a hybrid electric-hydraulic design: electric joints for most DOF with hydraulic actuation for the heaviest knee and hip joints. This hybrid philosophy influenced subsequent design practice.

Post-DRC Electric Transition (2016–2024)

The 2016–2024 period saw systematic migration from hydraulic to electric in academic and pre-commercial humanoid platforms, driven by four parallel developments: (1) high-energy-density NdFeB permanent magnets (Br > 1.3 T in 2024 production grades, versus 1.1 T in 2015) enabling higher motor torque constants; (2) silicon-carbide MOSFET inverters achieving 97–98% switching efficiency at 100 kHz switching frequency, reducing inverter cooling mass; (3) additive manufacturing enabling compact integrated joint actuator designs (motor + gearbox + encoder in 0.3–1.5 kg packages); (4) advances in quasi-direct-drive motor design (high pole count, concentrated windings, direct cooling) bringing electric joint power density to 1–3 kW/kg—competitive with hydraulic in the 100–500 W range per joint. Agility Robotics Cassie (2017) demonstrated dynamic bipedal walking with electric actuation and no stabilisation aid, achieving greater energy efficiency than comparable hydraulic platforms and demonstrating fall recovery—a milestone that shifted research community opinion on the viability of electric bipedal locomotion.

Tesla Optimus (announced 2022, Gen 2 released 2024) catalysed commercial investment in electric humanoid actuation at scale, triggering rapid iteration in motor controller and gearbox design across 20+ humanoid start-ups. Boston Dynamics’ announcement of electric Atlas in April 2024—alongside the retirement video of hydraulic Atlas performing gymnastics—confirmed that the electric crossover was complete for humanoid locomotion at the 50–100 kg body mass scale. The electric Atlas features 28 electric DOF with integrated rotary actuators using custom NdFeB motors, silicon-carbide FOC controllers, and quasi-direct-drive planetary transmissions; preliminary specifications indicate 2.5 kW/kg joint power density, approximately 40–60% of the hydraulic predecessor but sufficient for all locomotion and assembly tasks while eliminating the 8 kg HPU overhead and fluid maintenance requirement.

What Hydraulic Atlas Achieved That Electric Cannot Yet Match (2026)

As of 2026, three capabilities of the hydraulic Atlas generation remain unreplicated by electric humanoids: (1) Peak impact force absorption — hydraulic Atlas could absorb 3–5 kN impact forces at knee-joint during running without joint damage, owing to oil bulk modulus compliance and passive cylinder bottoming; electric counterparts with rigid gearboxes require explicit collision detection and rapid current limiting, which is slower. (2) Sustained heavy manipulation — hydraulic Atlas could exert 400 N pushing force at arm end-effector for 30+ seconds without thermal derating; electric arms with comparable reach reach winding temperature limits in 5–15 seconds at equivalent force. (3) Novel motion generation from compliance — the hydraulic Atlas parkour demonstrations (published December 2021) were partially enabled by the actuators’ natural energy storage during compression and release during explosive extension, a mechanism absent in electric systems without explicit spring elements.

Research and Literature

• Merritt, H.E. (1967). Hydraulic Control Systems. John Wiley & Sons — foundational second-order transfer function analysis and stability criteria for servo-hydraulic systems.
• Pratt, G.A. & Williamson, M.M. (1995). Series elastic actuators. Proc. IEEE/RSJ IROS, Pittsburgh PA, 399–406 — SEA compliance concept motivating hydraulic bulk-modulus force measurement.
• Semini, C. et al. (2011). Design of HyQ — a hydraulically and electrically actuated quadruped robot. Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering, 225(6), 831–849 — canonical hydraulic legged robot design paper.
• Jelali, M. & Kroll, A. (2003). Hydraulic Servo-systems: Modelling, Identification and Control. Springer — comprehensive model-based control treatment.
• Boaventura, T. et al. (2012). Model-based hydraulic impedance control for dynamic robots. IEEE Transactions on Robotics, 28(6), 1324–1336 — IIT HyQ impedance control framework.
• Seok, S. et al. (2015). Design principles for energy-efficient legged locomotion and implementation on the MIT Cheetah robot. IEEE/ASME Transactions on Mechatronics, 20(3), 1117–1129 — electric gear-ratio optimisation approaching hydraulic performance.
• Wensing, P.M., Wang, A., Seok, S. et al. (2017). Proprioceptive actuator design in the MIT Cheetah: Impact mitigation and high-bandwidth physical interaction. IEEE Transactions on Robotics, 33(3), 509–522 — QDD motor design demonstrating electric crossover.
• Semini, C. et al. (2019). HyQReal: Hydraulic quadruped with improved force and speed. IEEE Robotics and Automation Letters, 4(4), 4552–4559 — 3 tonne aeroplane tow demonstration.
• Tsagarakis, N.G. et al. (2017). WALK-MAN: A high-performance humanoid platform for realistic environments. Journal of Field Robotics, 34(7), 1225–1259 — hydraulic humanoid joint data.
• Focchi, M. et al. (2013). Robot impedance control and passivity analysis with inner torque and velocity feedback loops. Control Engineering Practice, 24, 97–112.
• Hutter, M. et al. (2016). ANYmal — a highly mobile and dynamic quadrupedal robot. Proc. IEEE/RSJ IROS — electric successor displacing hydraulic in legged research.
• Gehring, C. et al. (2016). Practice makes perfect: an optimisation-based approach to dynamic legged locomotion. Autonomous Robots, 40(3), 433–455.
• Kim, S. et al. (2021). MIT Mini Cheetah: A platform for pushing the boundaries on legged locomotion. IEEE Transactions on Robotics.
• Atkeson, C.G. et al. (2016). No falls, no resets: Reliable humanoid behavior in the DARPA robotics challenge. Proc. IEEE-RAS Humanoids — post-DRC hydraulic vs. electric analysis.
• Boston Dynamics (2024). Goodbye to HD Atlas. Press release, 16 April 2024. https://www.bostondynamics.com/atlas
• Bosch Rexroth (2022). Hydraulic Servo and Proportional Technology. Application guide RE 00293-01/02.2022. Lohr am Main.
• Moog Inc. (2023). Series 30 Servo Valve Product Data. Moog Controls Ltd, East Aurora NY.
• Moog Inc. (2023). Electrohydrostatic Actuators for Industrial Applications. Application white paper, Moog Industrial Division.
• Parker Hannifin (2024). Electro-Hydraulic Integrated Actuator (EHIA) — Technical Reference Manual. Parker Hannifin Hydraulics Europe Group.
• Sarcos Robotics (2021). Guardian XO Full-Body Powered Exoskeleton — Technical Specifications. Sarcos Technology and Robotics Corporation. https://www.sarcos.com
• ISO 4413:2010(E). Hydraulic Fluid Power — General Rules and Safety Requirements for Systems and Their Components. ISO/TC 131.
• NFPA T3.6.7:2014. Recommended Practice: Hydraulic Fluid Power — Systems — Performance Measurement Criteria. National Fluid Power Association.
• University of Bath PTMC Group (2023). Digital hydraulics for mobile machinery: efficiency, control, and thermal performance. International Journal of Fluid Power, 24(1), 1–28.
• Danfoss Power Solutions (2022). Digital Displacement® Pump: Efficiency and Controllability for Mobile Machinery. Technical white paper WP-DPPUMP-EN. Nordborg.
• UK Health and Safety Executive (2014). Safe Use of Hydraulic Equipment: HSG244. HSE Books, HMSO, Norwich.
• Siciliano, B. & Khatib, O. (2016). Springer Handbook of Robotics (2nd ed.), Chapter 7: Hydraulic and Pneumatic Actuation Systems. Springer, Berlin.
• Schilling Robotics (2023). Titan 4 Subsea Manipulator — Product Datasheet. Forum Energy Technologies, Houston TX.

Metadata

• Ontology anchor: robotics:HydraulicActuator — domain robotics confirmed correct from stub; no domain correction applied.
• IRI/URI: Retained http://narrativegoldmine.com/robotics#HydraulicActuator and urn:visionclaw:concept:robotics:hydraulic-actuator — consistent with stub and confirmed domain.
• Legacy term ID: RB-9011 — existing stub value retained; RB = Robotics domain prefix.
• Version: bumped 2.0.0 → 2.1.0 for Phase 6 enrichment.
• Domain correction: null — no correction required.
• Key 2024 event: Boston Dynamics hydraulic Atlas retirement 16 April 2024; electric humanoid transition fully documented.
• OWL axiom count: 43 SubClassOf axioms across 5 families: Compositional (9), Dependency (11), Capability (12), Implementation (11), Reduction (7); plus Data Properties (9), Property Constraints (4), Annotations (4), Property Characteristics (8).
• Wikilinks: 72 unique wikilink relationship instances across 11 relationship types.
• References in Provenance: 26 academic/industry/specification citations.
• Cross-domain bridges: Digital Twin, AI Agent System, Legged Robotics, Industrial Automation.

Provenance

• Merritt, H.E. (1967). Hydraulic Control Systems. John Wiley & Sons.
• Pratt, G.A. & Williamson, M.M. (1995). Series elastic actuators. IEEE/RSJ IROS, 399–406.
• Semini, C. et al. (2011). Design of HyQ. Proc. IMechE Part I, 225(6), 831–849.
• Jelali, M. & Kroll, A. (2003). Hydraulic Servo-systems: Modelling, Identification and Control. Springer.
• Boaventura, T. et al. (2012). Model-based hydraulic impedance control. IEEE T-Robotics, 28(6), 1324–1336.
• Seok, S. et al. (2015). Design principles for energy-efficient legged locomotion. IEEE/ASME T-Mechatronics, 20(3), 1117–1129.
• Wensing, P.M. et al. (2017). Proprioceptive actuator design in the MIT Cheetah. IEEE T-Robotics, 33(3), 509–522.
• Semini, C. et al. (2019). HyQReal. IEEE Robotics & Automation Letters, 4(4), 4552–4559.
• Tsagarakis, N.G. et al. (2017). WALK-MAN humanoid platform. Journal of Field Robotics, 34(7), 1225–1259.
• Focchi, M. et al. (2013). Robot impedance control. Control Engineering Practice, 24, 97–112.
• Hutter, M. et al. (2016). ANYmal. IEEE/RSJ IROS.
• Gehring, C. et al. (2016). Practice makes perfect. Autonomous Robots, 40(3), 433–455.
• Kim, S. et al. (2021). MIT Mini Cheetah. IEEE T-Robotics.
• Atkeson, C.G. et al. (2016). No falls, no resets. IEEE-RAS Humanoids.
• Boston Dynamics (2024). Goodbye to HD Atlas. Press release, 16 April 2024.
• Bosch Rexroth (2022). Hydraulic Servo and Proportional Technology. RE 00293-01.
• Moog Inc. (2023). Series 30 Servo Valve Product Data. Moog Controls Ltd.
• Moog Inc. (2023). Electrohydrostatic Actuators for Industrial Applications. White paper.
• Parker Hannifin (2024). EHIA Technical Reference Manual.
• Sarcos Robotics (2021). Guardian XO Technical Specifications.
• ISO 4413:2010(E). Hydraulic Fluid Power Safety Requirements. ISO.
• NFPA T3.6.7:2014. Hydraulic Fluid Power Performance Criteria. NFPA.
• University of Bath PTMC Group (2023). Digital hydraulics for mobile machinery. International Journal of Fluid Power, 24(1), 1–28.
• Danfoss Power Solutions (2022). Digital Displacement® Pump White Paper.
• UK HSE (2014). Safe Use of Hydraulic Equipment: HSG244. HSE Books.
• Siciliano, B. & Khatib, O. (2016). Springer Handbook of Robotics (2nd ed.). Springer.
• domain-correction: null — domain robotics confirmed correct; IRI/URI/same-as/owl-class unchanged
• legacy-term-id-note: RB-9011 retained from stub