Mobile Robot Platform

A Mobile Robot Platform is an integrated mechatronic base that provides locomotion, power, computing, and sensor-mounting infrastructure upon which higher-level autonomy stacks—perception, planning, and control—are deployed.

Semantic Classification

Content

Compositional Relationships (Components)

SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:LocomotionSystem))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:PowerSystem))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:OnboardComputer))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:SensorSuite))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:CommunicationInterface))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:MotorController))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:ChassisFrame))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:hasPart rb:PayloadBay))

## Dependency Relationships
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:requires rb:ROS2))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:requires rb:Nav2))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:requires rb:SLAM))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:requires rb:MotorDriver))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:requires rb:BatteryManagementSystem))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:requires rb:DDSMiddleware))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:dependsOn rb:LIDAR))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:dependsOn rb:IMU))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:dependsOn rb:DepthCamera))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:dependsOn rb:NVIDIAJetson))

## Capability Relationships
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:enables rb:AutonomousNavigation))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:enables rb:Teleoperation))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:enables rb:MobileManipulation))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:enables rb:EnvironmentMapping))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:enables rb:InspectionRobotics))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:enables rb:AgriculturalRobotics))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:enables rb:LogisticsAutomation))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:supports rb:MultiRobotSystems))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:supports rb:HumanRobotInteraction))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:supports rb:FleetManagement))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:supports rb:Sim2RealTransfer))

## Implementation Relationships
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:implements rb:SimultaneousLocalisationAndMapping))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:implements rb:PathPlanning))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:implements rb:SensorFusion))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:implements rb:BehaviourTreeControl))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:implements rb:LeggedLocomotion))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:implements rb:DifferentialDriveKinematics))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:uses rb:Nav2))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:uses rb:MoveIt2))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:uses rb:SLAMToolbox))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:uses rb:FastDDS))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:uses rb:BehaviourTree))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:uses rb:PointCloudLibrary))

## Reduction Relationships
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:reduces rb:HardwareDevelopmentTime))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:reduces rb:DeploymentCost))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:reduces rb:HumanExposureToHazard))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:reduces rb:InspectionDowntime))

## Association Relationships
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:relatedTo rb:ReinforcementLearning))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:relatedTo rb:ComputerVision))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:relatedTo rb:DigitalTwin))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:relatedTo rb:EdgeAI))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:contrastsWith rb:FixedRobotManipulator))
SubClassOf(rb:MobileRobotPlatform
  ObjectSomeValuesFrom(rb:contrastsWith rb:AerialRobot))

## Data Properties (Characteristics)
DataPropertyAssertion(rb:hasIdentifier rb:MobileRobotPlatform "RB-9017"^^xsd:string)
DataPropertyAssertion(rb:authorityScore rb:MobileRobotPlatform "0.87"^^xsd:decimal)
DataPropertyAssertion(rb:locomotionModalityCount rb:MobileRobotPlatform "5"^^xsd:integer)
DataPropertyAssertion(rb:turtleBot4UnitsDeployed rb:MobileRobotPlatform "18000"^^xsd:integer)
DataPropertyAssertion(rb:spotEnduranceMinutes rb:MobileRobotPlatform "90"^^xsd:integer)
DataPropertyAssertion(rb:unitreeGo2PriceUSD2024 rb:MobileRobotPlatform "1600"^^xsd:integer)

## Property Constraints
SubClassOf(rb:MobileRobotPlatform
  DataMinCardinality(1 rb:hasLocomotionModality xsd:string))
SubClassOf(rb:MobileRobotPlatform
  DataMinCardinality(1 rb:hasPayloadKg xsd:decimal))
SubClassOf(rb:MobileRobotPlatform
  DataAllValuesFrom(rb:hasROS2Support xsd:boolean))
SubClassOf(rb:MobileRobotPlatform
  DataSomeValuesFrom(rb:hasEnduranceMinutes xsd:integer))

## Annotations
AnnotationAssertion(rdfs:label rb:MobileRobotPlatform "Mobile Robot Platform"@en)
AnnotationAssertion(rdfs:comment rb:MobileRobotPlatform "Integrated mechatronic base providing locomotion, power, computing, and sensor-mounting infrastructure for autonomous mobile robotic systems. Encompasses wheeled (TurtleBot 4, Clearpath Husky/Jackal/Boxer), legged (Boston Dynamics Spot v3, Unitree Go2/B2/H1, MIT Mini Cheetah), and mobile-manipulator (Stretch RE3, Franka Panda on mobile base) platforms running ROS 2 Humble/Iron with Nav2, MoveIt 2, and SLAM Toolbox. Deployed across logistics, inspection, agriculture, and healthcare; key UK research centres include National Robotarium Heriot-Watt, ORI Oxford, Manchester, and Sheffield."@en)
AnnotationAssertion(dcterms:identifier rb:MobileRobotPlatform "RB-9017"^^xsd:string)
AnnotationAssertion(dcterms:subject rb:MobileRobotPlatform "Robotics, Autonomous Navigation, ROS 2, Legged Robots, Mobile Manipulation"@en)

)

Property Characteristics

AsymmetricObjectProperty(rb:requires) AsymmetricObjectProperty(rb:enables) AsymmetricObjectProperty(rb:implements) AsymmetricObjectProperty(rb:contrastsWith) TransitiveObjectProperty(rb:dependsOn) FunctionalDataProperty(rb:hasIdentifier) FunctionalDataProperty(rb:hasLocomotionModality)

Compositional Relationships

• A Mobile Robot Platform is composed of a Locomotion System providing mechanical movement via wheels, tracks, or legs; a Power System (lithium-ion or LiFePO4 battery pack with BMS) converting stored chemical energy into electrical power for all subsystems; an Onboard Computer stack executing the navigation, perception, and control software stack; a Sensor Suite integrating LIDAR, RGB-D cameras, IMU, and domain-specific sensors; a Communication Interface (Wi-Fi 6, 5G, ROS 2 DDS) for teleoperation, fleet management, and remote monitoring; a Motor Controller implementing real-time low-level control loops at 1 kHz; a Chassis Frame providing structural integrity, vibration isolation, and IP-rated environmental sealing; and a Payload Bay providing mechanical mounting (MIL-SPEC mounting holes, Picatinny rail, UR5 adapter flanges) and power/data interfaces (USB 3.0, Ethernet, CAN bus) for research payloads. These components are not merely additive — the kinematic design of the Locomotion System directly constrains the Power System capacity (legged platforms requiring 200–400 W locomotion vs 50–100 W wheeled), which cascades into Onboard Computer power budget and Sensor Suite selection.

Dependency Relationships

• Mobile Robot Platform requires ROS 2 as the middleware layer providing DDS-based publish-subscribe communication between Onboard Computer processes; requires Nav2 for autonomous waypoint navigation and obstacle avoidance in 2D/3D environments; requires SLAM (implemented via SLAM Toolbox or ORB-SLAM3) for simultaneous self-localisation and environment mapping when GPS is unavailable; requires Motor Driver hardware (typically CAN bus–connected servo drives or PWM motor controllers) to translate Nav2 velocity commands into wheel torques or joint torques; requires Battery Management System to protect lithium cells from over-discharge, over-temperature, and enable state-of-charge estimation; and requires DDS Middleware (FastDDS or CycloneDDS) to handle Quality-of-Service profiles for safety-critical real-time message delivery. The dependency on NVIDIA Jetson (or equivalent ARM+GPU compute) reflects the contemporary requirement for onboard neural network inference at 20–40 TOPS to run Computer Vision perception pipelines in real time without offloading to cloud.

Capability Relationships

• Mobile Robot Platform enables Autonomous Navigation (waypoint following, coverage, exploration) by providing the sensing-computing-actuation loop with Nav2 behaviour trees; enables Teleoperation by exposing geometry_msgs/Twist ROS 2 topics to remote joystick or web interface operators; enables Mobile Manipulation when paired with a manipulator arm (Franka Panda, UR5, Spot Arm) via whole-body MoveIt 2 motion planning; enables Environment Mapping through LIDAR-based SLAM producing 2D occupancy grids or 3D point-cloud maps; enables Inspection Robotics by navigating to pre-programmed inspection waypoints and triggering sensor acquisition; enables Agricultural Robotics through GPS-guided row-following and crop-proximity sensing; and enables Logistics Automation through LIDAR-based people-avoidance AMR navigation in warehouse environments. The platform also supports Multi-Robot Systems via ROS 2 namespacing and multi-master DDS domain configuration, supports Human Robot Interaction through proxemics-aware navigation and voice/gesture interfaces, and supports Sim2Real Transfer by exposing the same ROS 2 topics in Gazebo Harmonic or Isaac Sim simulation as on physical hardware.

Implementation Relationships

• Mobile Robot Platform implements Simultaneous Localisation and Mapping via SLAM Toolbox (Karto scan-matching backend, loop-closure detection, lifelong map management); implements Path Planning via Nav2 GlobalPlanner (NavFn/Theta*/Smac 2D/3D) and LocalPlanner (DWB or MPPI); implements Sensor Fusion via robot_localization EKF/UKF fusing wheel odometry, IMU, and GPS into a filtered /odom → /map TF tree; implements Behaviour Tree Control via the BT Navigator with custom-designed XML behaviour trees for mission execution; implements Legged Locomotion (on quadruped/biped platforms) via model predictive control (MPC) or learned RL policies executing at 1 kHz on real-time controllers; and implements Differential Drive Kinematics (on wheeled platforms) via the diff_drive_controller from ros2_control computing wheel velocities from Twist commands. The PointCloud Library (PCL) implements 3D point cloud filtering, clustering, and surface normal estimation for obstacle representation in 3D costmaps.

Reduction Relationships

• Mobile Robot Platform reduces Hardware Development Time by providing validated, pre-integrated mechanical, electrical, and software assemblies eliminating 12–24 months of custom platform development; reduces Deployment Cost by lowering the total cost of ownership versus custom UGV builds ($20,000–$80,000 commercial platform vs $150,000–$400,000 custom bespoke); reduces Human Exposure to Hazard by performing inspection and data collection tasks in nuclear, chemical, offshore, and confined-space environments inaccessible or dangerous for humans; and reduces Inspection Downtime by enabling continuous autonomous monitoring (24/7 with docking-based autonomous recharging) versus periodic manual inspection cycles (weekly/monthly).

About Mobile Robot Platforms

• Mobile Robot Platforms are integrated mechatronic systems providing the physical and computational foundation for autonomous or teleoperated mobile robotic applications. They abstract away the significant engineering effort of designing locomotion mechanics, power management, motor controllers, and embedded computing, allowing researchers and engineers to focus on higher-level autonomy—perception, planning, learning, and human-robot interaction. The concept encompasses a vast continuum: from the $200TurtleBot3usedtoteachglobalundergraduatecohorts,tothe$75,000 Boston Dynamics Spot equipped with an Orbit software suite managing multi-site inspections for BP, EDF, and National Grid.
• The field has undergone two transformative shifts since 2020. First, ROS 2 (Robot Operating System 2) replaced ROS 1 as the dominant middleware, providing real-time DDS communication, improved security, lifecycle-managed nodes, and production-quality deployment capabilities. Second, learned locomotion policies via Reinforcement Learning displaced hand-engineered analytical controllers for legged robots: Unitree’s RL framework and MIT’s agility research demonstrated quadrupeds traversing rubble, stairs, and outdoor terrain that previously required careful manual tuning of spring-mass models. These two shifts accelerated commercialisation dramatically. The global mobile robot market exceeded $22billionUSDin2024andisprojectedtoreach$58 billion by 2030 (MarketsandMarkets, 2025), driven by logistics, inspection, agriculture, and healthcare verticals.
• The underlying engineering discipline integrates insights from kinematics and dynamics (rigid body motion, Denavit-Hartenberg parameters for manipulator chains, Newton-Euler recursive dynamics for legged robots), estimation theory (Kalman filtering, particle filters, factor graph optimisation for SLAM back-ends), control theory (PID for wheel velocity, model predictive control for trajectory execution, whole-body QP controllers for legged platforms), and machine learning (convolutional neural networks for 3D object detection, transformer-based vision-language-action models, and policy gradient methods for locomotion). The integration of these disciplines within a single deployable mechatronic system—operating outdoors, in real time, under tight power and compute budgets—is what distinguishes mobile robot platform engineering from any single constituent discipline.

Components and Architecture

Every mobile robot platform integrates five functional layers that must be carefully co-designed because the constraints propagate bidirectionally: the mass of the payload affects locomotion power demand, which affects battery size, which affects chassis structural requirements, which feeds back into total mass. The co-design problem is typically framed as a multi-objective optimisation over endurance (Wh/kg of payload), terrain capability (maximum slope, step height, ground clearance), and cost (USD/kg payload).

Platform selection taxonomy by locomotion modality:

Modality	Best for	Max slope	Step height	Example platforms
Differential wheeled	Flat/indoor research	10°	0 cm	TurtleBot 4, Jackal
Skid-steer	Outdoor unstructured	35°	15 cm	Husky A200, Boxer
Mecanum omnidirectional	Indoor precision	5°	0 cm	Ridgeback, Boxer
Quadruped legged	Stairs/rubble	45°	30 cm	Spot v3, Go2 Pro
Tracked	Extreme terrain	60°	40 cm	Grizzly, custom UGV
Biped humanoid	Human environments	30°	25 cm	H1, G1, Atlas
Mobile manipulator	Manipulation + move	10°	0–30 cm	Stretch RE3, Panda+Ridgeback

Platform selection heuristics (2024-2025):

• For flat indoor environments requiring precise positioning (±5 cm): wheeled omnidirectional (Clearpath Boxer) or differential drive (TurtleBot 4). Nav2 AMCL with 2D LIDAR achieves 3–5 cm localisation RMS.

• For outdoor unstructured terrain with slopes up to 15°: skid-steer (Clearpath Husky A200). SLAM Toolbox 3D with Ouster OS1 achieves 10–20 cm localisation RMS on grass.

• For stair and step climbing (up to 30 cm step height): quadruped legged (Spot v3, Unitree Go2 Pro). RL-based locomotion policy generalises to novel geometry without re-training.

• For manipulation + mobility: mobile manipulator (Stretch RE3 for lightweight domestic; Ridgeback + Franka Panda for precision research; Spot + Arm for outdoor dexterous).

• For humanoid manipulation in human-structured environments: biped (Unitree H1/G1, Atlas Electric 2025) executing whole-body imitation learning policies.

  1. Locomotion Subsystem

  The mechanical foundation determines the robot’s kinematic envelope and terrain capability. Differential-drive wheeled platforms (two driven wheels, one or two passive casters) provide the simplest kinematics—velocity commands map directly to left/right wheel speeds via v = r(ω_L + ω_R)/2, ω = r(ω_R − ω_L)/L—but cannot move sideways. Omnidirectional platforms (Mecanum or holonomic wheels) achieve full planar mobility at the cost of reduced traction and load capacity. All-terrain skid-steer platforms (Clearpath Husky A200, Boxer) drive all four or six wheels and steer by differential torque, tolerating grass, gravel, and moderate slopes at the cost of energy-intensive turning. Legged platforms unlock vertical mobility (stairs, ladders, rough terrain) through model predictive control or learned locomotion policies; at the cost of mechanical complexity, energy consumption, and payload-to-mass ratios of 0.2–0.5 versus 1–3 for wheeled platforms.

  Key specifications by category (2024-2025):

• Clearpath Jackal UGV: 12 kg, 2 m/s, 3 h endurance, 10 kg payload, IP54, ROS 2 native. Preferred indoor-outdoor research platform at Edinburgh Robotarium and ORI Oxford.

• Clearpath Husky A200: 50 kg, 1 m/s, 3 h, 75 kg payload, IP54. De-facto large outdoor research standard; 400+ deployments globally including NRL Heriot-Watt (8 units) and Autonomous Mobile Robotics Lab Manchester.

• Clearpath Boxer: 65 kg, 2 m/s, 8 h, 60 kg payload, IP52. Omni-directional indoor industrial platform.

• Boston Dynamics Spot v3 (2024): 32 kg, 1.6 m/s, 90 min, 14 kg payload, IP54. Arm optional ($15Kadd-on).Orbitenterprisesoftware$1,450/month subscription (2024). 4,000+ units deployed globally.

• Unitree Go2 Air (2024): 4 kg, 3.5 m/s, 40 min, 5 kg payload, IP67. Consumer/education tier, $1,600 USD. 15,000+ units shipped by Q4 2024.

• Unitree B2 (Q2 2024): 60 kg, 1.5 m/s, 5 h, 20 kg payload, IP67, ATEX Zone 2 option. Industrial inspection target.

• Unitree H1 Humanoid (Q4 2024): 47 kg biped, 1.5 m/s walking, 10 kg carried load, 3 DOF per leg, full-body whole-body control. $90,000 USD research price.

• Unitree G1 (Q2 2025): 35 kg biped, 127 cm, 7-DOF arms, $16,000 USD. Targeting manipulation research.

• MIT Mini Cheetah (open-source): 9 kg, 3 m/s, 3 DOF per leg, 12 actuated joints, open-source Cheetah-Software and hardware BOM. Foundation for 50+ academic legged-robot research programmes.

• Hello Robot Stretch RE3 (2023): 25 kg, 1.2 m mast with telescoping arm, 7-DOF manipulation, ROS 2 Humble, Python SDK. $23,000 USD. Targeted at assistive and elder-care research.

• Franka Emika Panda on mobile base: 7-DOF, 3 kg payload arm mounted on omnidirectional base (e.g., Clearpath Ridgeback, 100 kg, 0.4 m/s); yields fully capable mobile manipulator for research at ICL, UCL, Sheffield, TRI.

  2. Power and Energy Management

  Lithium-ion and LiFePO4 battery packs (24–48 V, 10–40 Ah) power the platform with onboard battery management systems (BMS) monitoring cell voltages (±10 mV accuracy), state-of-charge estimation via Coulomb counting + Kalman filter, and thermal cutoffs (>60°C). Endurance is the primary commercial constraint: 40–90 minutes for legged platforms versus 3–8 hours for wheeled. Regenerative braking (Clearpath, Boxer) recovers 5–15% energy on ramps. Fast-charging stations (Boston Dynamics Spot Dock, Unitree Go2 dock) enable autonomous 30-minute recharge cycles for 24/7 operation. Power budgeting: locomotion typically consumes 50–200 W, onboard computers 20–80 W, sensors (LIDAR + cameras) 10–30 W, payload 0–200 W.

  3. Onboard Computing Infrastructure

  The compute hierarchy consists of real-time motor controllers (ARM Cortex-M7, 1 kHz control loop), a navigation computer (NVIDIA Jetson Orin NX 16 GB at 40 TOPS for Nav2 + SLAM + perception inference, consuming 25 W), and optionally a second high-power GPU node (Jetson AGX Orin 64 GB at 275 TOPS, or discrete NVIDIA RTX for point-cloud deep learning). ROS 2 Iron (long-term support through 2027) and Humble (LTS through 2027) are the dominant middleware releases as of 2025. FastDDS is the default DDS implementation; CycloneDDS sees preference in latency-sensitive manipulation applications.

  4. Sensor Suite

  Standard research platforms mount:

• LIDAR: Ouster OS1/OS2 (32–64 channel, 120 m range, 320K points/s), Velodyne VLP-16 (legacy), Livox MID-360 (solid-state, 360° coverage, <$1,000). Primary input to SLAM and Nav2 costmaps.

• RGB-D Camera: Intel RealSense D435i (depth 0.3–10 m, 30 fps, IMU integrated) for point-cloud augmentation and 3D object detection.

• IMU: VectorNav VN-100 or Xsens MTi-30 (9-DOF, 100–400 Hz) for odometry fusion and state estimation.

• RGB Cameras: Intel RealSense colour stream or dedicated USB cameras for visual SLAM (ORB-SLAM3, Kimera), apriltag localisation, and semantic perception.

• Spot-specific: Boston Dynamics Spot v3 ships five fisheye cameras (360° coverage), one depth camera, and optional arm-mounted Pan-Tilt-Zoom camera and Spot CAM+ (360° colour + thermal).

• Unitree Go2: Forward stereo depth camera, LIDAR optional add-on ($800 USD Unitree L1 LIDAR).

  5. Software Architecture

  The canonical ROS 2 software stack for a mobile robot platform comprises:

• SLAM Toolbox: Lifelong mapping, online SLAM, graph-based pose optimisation (Karto backend). Replaces legacy AMCL for map-building; AMCL retained for pure localisation in known maps.

• Nav2: ROS 2 navigation stack. Costmap 2D with inflation, obstacle, voxel, and range layers; BT Navigator with lifecycle-managed behaviour trees; Controller server (DWB planner or MPPI controller—added Nav2 Humble); Smoother server; Recovery server (spin, wait, backup). Demonstrated autonomous 99.97% task completion across >50,000 navigation episodes in the Edinburgh Robotarium testbed (2024).

• MoveIt 2: Manipulation planning, collision-aware Cartesian motion planning (STOMP, OMPL, Pilz Industrial), servo (real-time jogging), MoveIt Task Constructor for multi-step assembly tasks.

• micro-ROS: Runs ROS 2 directly on microcontrollers (STM32, ESP32) for motor controllers and sensor nodes; eliminates serial bridge latency.

• Gazebo: Physics-accurate simulation for Nav2 and MoveIt development. Gazebo Harmonic (November 2023) unified the Ignition Gazebo branding and supports domain randomisation for reinforcement learning training loops.

• NVIDIA Isaac Sim: High-fidelity GPU-accelerated photorealistic simulation (Omniverse physics, RTX ray tracing) for sim2real transfer of perception models. Isaac ROS 2 bridges provide drop-in compatibility with Nav2 and MoveIt 2.

  ROS 2 QoS and real-time considerations: Navigation-critical topics (/cmd_vel, /joint_states, /tf) use RELIABLE reliability with KEEP_LAST depth-1 QoS profiles for guaranteed delivery; sensor topics (/scan, /camera/image_raw) use BEST_EFFORT VOLATILE to prioritise latency over reliability. The DDS Middleware (FastDDS with shared-memory transport, CycloneDDS with Zero-Copy) achieves 50–200 µs intra-process latency for the motor control loop vs 1–5 ms over the network interface — critical for stable PID loops at 1 kHz. The ros2_control framework abstracts the hardware interface (PositionJointInterface, VelocityJointInterface, EffortJointInterface) enabling the same Nav2/MoveIt 2 software stack to run across Husky (velocity-controlled wheel motors), Stretch RE3 (Dynamixel position-controlled servos), and Spot (proprietary joint-torque API bridged via ros2_control hardware plugin). This portability is the primary value proposition of ROS 2 for the mobile robot platform ecosystem: write navigation logic once, deploy across ten platform types with only a hardware plugin swap.

  Security and production deployment: ROS 2 Security (SROS2) provides mutual TLS authentication and AES-256 encryption for DDS communications, critical for industrial deployments where robot commands must be authenticated. SROS2 is mandatory in Spot Enterprise deployments (Boston Dynamics Orbit cloud integration). The ros2_lifecycle node lifecycle state machine (unconfigured → inactive → active → finalized) enables graceful startup sequencing (sensors initialise before navigation stack activates), preventing the race conditions common in ROS 1 deployments.

  Digital Twin integration: Nav2 and Isaac Sim 4.x both publish ROS 2 tf2 transform trees readable by the same monitoring dashboards. Spot Orbit (enterprise fleet management) and Clearpath RMF (robotics middleware framework) integrate ROS 2 navigation topics with facility management systems (BIM, CMMS) providing live robot position overlays on CAD floor plans and automated work order generation on inspection anomaly detection.

ROSCon 2024–2025 Highlights and Open Robotics Releases

ROSCon is the premier annual community conference for ROS developers. ROSCon 2024 (Odense, Denmark, October 2024) and ROSCon 2025 (Kyoto, Japan, October 2025) featured key announcements relevant to mobile robot platforms:

ROSCon 2024:

• Open Robotics announced ROS 2 Kilted Kaiju (planned May 2025 release, 5-year LTS), with major improvements to ros2_control (distributed controllers across networked compute nodes) and Nav2 (probabilistic semantic costmaps using 3D detection results from perception pipelines).

• Boston Dynamics released Spot SDK 4.0 with enhanced GraphNav 2.0 (visual loop closure on texture-sparse industrial environments), Spot CAM+ thermal anomaly detection, and Core I/O compute payload now supporting Docker containers for custom ROS 2 applications.

• NVIDIA demonstrated Isaac ROS 2 Humble packages: cuVSLAM (GPU-accelerated visual SLAM 30× faster than CPU ORB-SLAM3), Isaac Perceptor (multi-camera 3D semantic segmentation at 30 fps on Jetson AGX Orin), and Isaac Manipulator (cuRobo trajectory generation at 2,000 trajectories/second for MoveIt 2 integration).

• Unitree Robotics presented Go2 Developer SDK (Python + C++ ROS 2 wrapper), enabling direct joint-level control from ROS 2 topics, triggering adoption at 200+ universities worldwide.

• University of Edinburgh presented results from the National Robotarium long-term autonomy testbed: 8 Husky A200 robots operated continuously for 30 days with <0.1% navigation failure rate using Nav2 Humble + SLAM Toolbox lifelong mapping.

• micro-ROS project released micro-ROS for Arduino enabling TurtleBot 4’s iRobot Create 3 base to run ROS 2 subscriber/publisher nodes at 1 kHz for wheel odometry.

  ROSCon 2025:

• Open Robotics released Nav2 with Coverage Navigation: complete area coverage planning (Boustrophedon decomposition, spiral, contour-following) targeting agricultural and floor-cleaning applications. Deployed on Thorvald and Temi platforms.

• Seven humanoid platforms demonstrated live ROS 2 integration: Unitree G1, Figure-02, 1X Neo, Agility Robotics Digit v4, Apptronik Apollo, Sanctuary Phoenix 2, and Kepler K1. All used MoveIt 2 for arm planning; locomotion ranged from custom RL policies to Boston Dynamics Atlas Electric’s hydraulic-free proprioceptive MPC.

• Zenoh bridge for ROS 2 (rmw_zenoh_cpp 1.0) released as production-stable, enabling cloud-connected ROS 2 deployments with 99.99% uptime across NAT firewalls — enabling fleet management for Spot deployments at remote offshore sites.

• Edinburgh Robotarium released open-source multi-robot Nav2 stack for heterogeneous fleets (Husky + Jackal + TurtleBot 4 operating collaboratively via shared costmap over ROS 2 DDS).

Use Cases and Major Platform Families

Logistics and Warehousing

Boston Dynamics Stretch—a wheeled mobile base with a 6-DOF arm and suction gripper—achieved commercial deployment in 2024 at DHL and Gap Inc. warehouses, palletising/depalletising at 800 boxes/hour with 99.9% pick success on mixed-SKU cartons. The Stretch base uses four omnidirectional wheels and a counterbalanced arm that can handle cartons 5–23 kg from conveyors at heights 0.2–2.0 m. Clearpath OTTO 100 (100 kg payload) and OTTO 1500 (1,500 kg payload) AMRs navigate without floor infrastructure modifications using LIDAR SLAM (SLAM Toolbox 3D), deployed in 850+ factories and distribution centres globally. Six River Systems (Shopify acquisition) deployed 12,000+ wheeled AMRs in Shopify merchant fulfilment centres. Unitree B2 pilots conveyor-line autonomous visual inspection with mounted RGB-D camera. The logistics automation market reached $18Bin2024withAMRscapturing34$42B by 2029 (IDC, 2025).

Key UK logistics deployments:

• ASOS Barnsley distribution centre: Ocado Automated Storage and Retrieval System with 3,000+ bespoke wheeled robots on 3D grid — UK’s largest autonomous warehousing deployment.

• Ocado Technology, Hatfield: In-house-developed 4-wheel omnidirectional robots (Ocado Smart Platform, OSP) processing 800,000+ orders/week. ROS-based, proprietary navigation stack.

• Amazon UK fulfilment centres (Coventry, Manchester, Swansea): 1,200+ Proteus AMRs (Amazon Robotics) for pod-to-picker inventory movement.

• DHL UK Innovation Centre, Coventry: Spot inspection pilot + Locus Robotics AMR fleet for goods-to-person picking.

  Industrial Inspection

  Boston Dynamics Spot with Spot CAM+ and Core I/O payload conducts gas detection (Honeywell gas sensor payload), thermal imaging (FLIR Lepton integration), and pressure gauge reading (computer vision) at 200+ oil, gas, and chemical facilities globally.

  Key UK and European deployments:

• BP Valhall offshore platform: Spot teleoperated from Aberdeen onshore control room via satellite link (100–300 ms latency), conducting daily inspection rounds replacing weekly helicopter visits. Saves £120,000/year in helicopter costs and eliminates offshore worker exposure.

• National Grid UK electricity substations (15+ sites): Spot conducting autonomous infrared thermography inspection of high-voltage switchgear, detecting hotspots indicating insulation failure before outage. Operating in 400 kV substations where human access requires switching operations and 4-hour permit procedures.

• EDF Energy Hinkley Point C nuclear construction site (2023–2026): ANYmal-D (ANYbotics, Heriot-Watt Robotics Institute collaboration) conducting visual inspection of rebar placement and concrete pour quality in high-radiation zones inaccessible to humans during active pours.

• Sellafield nuclear reprocessing, Cumbria: Clearpath Husky A200 with radiation survey payload (RadEye SPRD) mapping contamination distribution in legacy waste stores, ORI Oxford IM-BRAIN project 2023–2025.

  ANYbotics ANYmal-D specifications relevant to inspection: IP67, ATEX Zone 2 certified (IECEx SIR 22.0030), -20°C to 55°C operating range, 90 min endurance, Spot-equivalent autonomous navigation, acoustic leak detection payload (ultrasonic microphone array). Used at 50+ oil/gas facilities globally.

  Agricultural Robotics

  University of Lincoln Thorvald robot (4-wheel omnidirectional, ROS 2 Humble, Velodyne VLP-16 LIDAR) achieves strawberry harvesting at 12,000 berries/day versus 17,000 for a skilled human picker, operating 20 h/day on two automated charge cycles, at 70% of human labour cost. The Thorvald platform (Saga Robotics spin-out) is now commercially deployed at Dyson Farming (Lincolnshire) and Driscoll’s berry farms (California).

  Wider UK agricultural robotics platform deployments:

• Hands Free Hectare, Harper Adams University (Shropshire): First fully autonomous arable farm cycle (2018) using Clearpath Jackal for scouting and Husky A200 for weed mapping; extended to multi-season continuous autonomous farming trials 2019–2024.

• Agri-Epi Centre at Harper Adams: Clearpath Jackal fleet with Micasense Altum-PT multispectral cameras (5 spectral bands + thermal) for NDVI/LAI crop monitoring across 400 ha demonstration farm.

• Small Robot Company (Wilton, Wiltshire): Tom (survey robot, 500 g, ultrasonic + camera) and Dick (herbicide micro-dosing robot, 20 kg) for precision per-plant intervention, ROS 2 based. Partnership with Dyson Farming for 2024–2025 UK wheat trials covering 8,000 ha.

• FieldWork Robotics (Bristol, University of the West of England spin-out): Raspberry and mushroom harvesting robots on differential-drive bases, ROS 2, 4× LIDAR for row navigation, neural network fruit detection (YOLOv8) at 94% recall.

  Healthcare and Assistive Robotics

  Hello Robot Stretch RE3 pilots at Stanford CHARM Lab (elder-care tasks: medication delivery, surface wiping, object retrieval for motor-impaired users), MIT CSAIL (whole-body manipulation, MoveIt Task Constructor), and University of Sheffield ACSE (assistive manipulation for motor-impaired users under EPSRC “Assistive Robots for Independent Living” grant 2024–2027). The Stretch RE3 compact form factor (34 cm × 34 cm footprint), 25 kg mass, and $23,000 USD price enable NHS-scale deployment trials absent from heavier 65–80 kg alternatives. Key technical attributes for clinical environments: 1.2 m telescoping mast (reaching high shelves without tipping), hook and dexterous gripper end-effectors (switchable in <60 s), ROS 2 Python navigation API enabling care staff to define task programs without robotics expertise.

  UK NHS healthcare deployments (2024–2026):

• Pepper (SoftBank Robotics, 120 cm biped): Wayfinding and appointment check-in at 50+ UK NHS Trusts including Sheffield Teaching Hospitals, Bristol Royal Infirmary, and King’s College Hospital London.

• Temi V3 (telepresence platform): Remote GP consultation and ward round support at Royal Free London NHS Foundation Trust (pilot 2024), enabling physicians to examine patients remotely during infectious disease outbreaks.

• Moxi (Diligent Robotics, Texas): Hospital logistics (specimen transport, linen delivery) piloting at UCLH (University College London Hospitals, 2025), 35 kg differential-drive, ROS 2, Nav2 with AMR-IT hospital integration layer.

  Field Robotics and Disaster Response

  MIT Mini Cheetah (open-source hardware BOM, Cheetah-Software GitHub) and University of Edinburgh extensions achieve 3 m/s running on flat terrain and 1.5 m/s on rubble with learned RL policies (Ji et al., IEEE RA-L 2022). The open-source release of Mini Cheetah hardware designs in 2023 enabled 200+ academic groups globally to build and study the platform, accelerating the RL locomotion research pace.

  DARPA Subterranean Challenge (SubT) 2021: winner Team CERBERUS (ETH Zurich, ANYmal-D × 2 + Scout Mini quadruped + aerial) demonstrated multi-robot autonomous subterranean exploration — navigating 2 km of unknown underground mine tunnels in 60 minutes without human waypoint guidance. Technical approach: graph-based exploration with SLAM (LIO-SAM lidar-inertial odometry, 3D occupancy mapping), multi-robot communication via mesh radio, and priority-based task assignment. The SubT Challenge directly influenced UK EPSRC ORCA Hub and IM-BRAIN programme specifications for autonomous mine inspection.

  Spot v3 Arm commercial deployment for defence and public safety:

• US Army 82nd Airborne Division (2024): Spot for forward reconnaissance, route clearing, and EOD scouting — 12 units fielded.

• NYPD (New York Police Department): Spot for stairwell search-and-clear, hostage situation reconnaissance; generated public controversy leading to usage policy review.

• UK Ministry of Defence: Defence and Security Accelerator (DASA) contract to Army AI Centre (Upavon) for Spot evaluation in counter-IED scouting, 2024.

  Research and Education

  TurtleBot 4 (released 2022, Clearpath Robotics + iRobot Create 3 base, ROS 2 Humble native) shipped 18,000+ units to universities globally by 2025. At £1,200–£1,900 GBP (Standard and Pro variants), it is the primary teaching platform for ROS 2 at Edinburgh, Manchester, UCL, Imperial, Sheffield, Cambridge, and Bristol. The Create 3 base runs micro-ROS natively, exposing wheel odometry and IMU via standard ROS 2 topics. TurtleBot 4 Pro adds an OAK-D Pro W depth camera and additional USB/Ethernet expansion ports. The National Robotarium at Heriot-Watt maintains 8 Husky A200s, 4 Jackals, 2 ANYmal-D units, and 1 Spot v3 in a 700 m² arena with Vicon Vantage 16-camera motion capture (sub-millimetre ground truth for SLAM evaluation) and an 8× NVIDIA A100 GPU compute cluster (supporting Isaac Lab and MuJoCo parallel RL training).

  Open-source platform ecosystem:

• TurtleBot 4: Primary ROS 2 education platform, 18,000+ units.

• MIT Mini Cheetah: Open-source legged research platform, 200+ academic builds post-2023 hardware release.

• Unitree Go2 EDU: $2,800 USD developer version with full joint-level API access, 5,000+ deployed at universities.

• Open Dynamic Robot Initiative (ODRI): ETH Zurich + MPI-IS modular legged robot hardware BOM (12-DOF quadruped for $3,000 USD in components), enabling fully open-source legged robot research.

Safety, Standards, and Regulation

Mobile robot platforms operating in public or mixed human-robot environments are subject to safety standards that constrain design and deployment:

ISO 13482:2014 “Safety requirements for personal care robots”: Mandates risk assessment for mobile servant robots (Type A) and physical assistant robots (Type B) operating in proximity to humans. Applies to Pepper, Temi, and Stretch RE3 in healthcare settings. Requires force/torque limits (< 50 N impact for non-hazardous contact), emergency stop within 200 ms, and geofencing capability.

ISO 3691-4:2020 “Industrial trucks — Safety requirements and verification — Part 4: Driverless industrial trucks and their systems”: Governs AMRs in logistics (Clearpath OTTO, Boston Dynamics Stretch warehouse deployment). Mandates lidar-based safety zones (protective field + warning field), failsafe braking (<1 s stopping from max speed), and documented SIL 2 (Safety Integrity Level 2) for safety-critical control paths.

EU Machinery Regulation 2023/1230 (replacing Machinery Directive 2006/42/EC): Extends risk assessment requirements to AI-driven autonomous machinery. Mobile robots with learned locomotion policies must document AI system validation procedures. Directly affects CE marking of Spot, ANYmal-D, and Unitree B2 for EU industrial deployment.

ATEX Directive 2014/34/EU: ANYbotics ANYmal-D ATEX Zone 2 certification (IECEx SIR 22.0030) permits operation in flammable gas atmospheres (offshore, chemical plant), requiring spark-ignition-resistant actuators, sealed battery enclosures, and intrinsic safety (Ex i) classification for all onboard electronics.

UK-specific: UKRI EPSRC established the Robots for a Safer World programme (2022-2026) to develop safety assurance frameworks for field robots in nuclear decommissioning, alongside the National Physical Laboratory developing measurement standards for robot performance benchmarking (NPL Good Practice Guide No. 167, “Measurement of Mobile Robot Navigation Performance”, 2024). HSE (Health and Safety Executive) issued guidance on Autonomous Mobile Robots in workplaces (INDG 2024/07) requiring documented human-machine interaction safety cases for AMR deployments exceeding 100 kg operating near workers.

Academic Context

Mobile robot platform research integrates contributions from control theory, mechanical engineering, computer vision, and machine learning.

Foundational Works

Siegwart, Nourbakhsh, and Scaramuzza’s Introduction to Autonomous Mobile Robots (2nd ed., MIT Press, 2011) remains the canonical graduate textbook. Thrun, Burgard, and Fox’s Probabilistic Robotics (MIT Press, 2005) established Bayesian estimation frameworks—particle filter AMCL, EKF-SLAM, occupancy grids—that underpin Nav2’s localisation pipeline to this day. Corke’s Robotics, Vision and Control (Springer, 2nd ed. 2017) provides the computational geometry and control background.

Legged Locomotion

Raibert’s Legged Robots That Balance (MIT Press, 1986) established the spring-loaded inverted pendulum (SLIP) model still referenced in Spot’s dynamic gait analysis. The SLIP model reduces the complex multi-DOF legged system to a mass-spring abstraction with a single passive compliance element, enabling analytical derivation of stable periodic gaits (trot, canter, bound) that informed the trajectory optimisation approaches (DDP, SQP) used in early Boston Dynamics BigDog and PETMAN platforms.

The modern Reinforcement Learning era in legged locomotion traces to Unitree, ETH Zurich (ANYbotics), and MIT Biomimetics Lab: Hwangbo et al. (Science Robotics 2019) demonstrated the first successful end-to-end learned locomotion policy on ANYmal (ETH), using an actuator network trained to replicate real-hardware dynamics in simulation — achieving 0.75 m/s on stairs unseen during training. The actuator network approach (predicting joint torque given position error, velocity, and previous torque) proved critical for bridging the sim2real gap caused by unmodelled actuator elasticity and thermal effects. Kumar et al. (RSS 2021) showed zero-shot sim2real transfer on Unitree A1 quadruped via Rapid Motor Adaptation (RMA): a base policy trained in simulation with domain randomisation, plus an online adaptation module (64-dimension extrinsics vector estimated from 50-step proprioceptive history) that adjusts to terrain parameters (friction, mass perturbation) in real time without any real-world training. Ji et al. (IEEE RA-L/ICRA 2022) demonstrated MIT Mini Cheetah running at 3.9 m/s on flat terrain with concurrent RL-based state estimation (terrain normal + contact events) trained alongside the locomotion policy. Zhuang et al. (Unitree, IROS 2024) released the first publicly available full RL-trained policy for Unitree Go2 achieving 3.5 m/s outdoor runs with recovery from lateral falls, trained in Isaac Sim with 4,096 parallel environments on 8× A100 GPUs for 48 hours.

Whole-body control (WBC): The alternative to end-to-end RL is hierarchical WBC separating high-level task specification (Cartesian end-effector target) from low-level torque computation via constrained QP. Bellicoso et al. (IROS 2019) demonstrated ANYmal WBC enabling simultaneous walking and arm operation. Schwarm et al. (ICRA 2023) combined WBC with Reinforcement Learning pre-training to accelerate convergence. The tension between end-to-end RL (simpler, more robust, harder to interpret) and WBC (interpretable, easier to specify, requires accurate dynamics model) remains unresolved; most production systems (Spot Arm, ANYmal-D) use WBC for safety-critical manipulation and RL for adaptive locomotion.

Navigation, SLAM, and Long-Term Autonomy

Macenski et al. (IROS 2020) introduced SLAM Toolbox — graph SLAM with Karto scan-matching, life-long mapping capability (incremental pose graph addition without full re-optimisation), and multi-session mapping (loading prior maps and extending them). SLAM Toolbox became the Nav2 default mapping backend in Humble, replacing GMapping/RTAB-Map for ROS 2 deployments. Macenski et al. (Robotics and Autonomous Systems, 2023) provided a comprehensive survey of Nav2 algorithms covering 12 global planner plugins and 8 local controller plugins.

The Model Predictive Path Integral (MPPI) controller (Williams et al. ICRA 2017, Georgia Tech) samples K=2,000 random control sequences in parallel on GPU, evaluates their costs using the robot dynamics model + costmap, and selects the optimal sequence via importance-sampling weighted average. MPPI was integrated into Nav2 Humble (2022) as the default local controller, replacing the Dynamic Window Approach (DWA). MPPI advantages: handles non-convex cost landscapes (narrow corridors, dynamic obstacles) without local minima, and naturally avoids dynamic obstacles at 10 Hz replanning frequency. ORB-SLAM3 (Campos et al. IEEE PAMI 2021) extended visual-inertial SLAM to support multi-map (handling revisits from different entry points), fisheye cameras (Spot’s five-camera ring), and IMU tight integration (pre-integration on manifold SO(3)), achieving <1% trajectory error on EuRoC MAV benchmark. Deployed on Stretch RE3 (monocular mode) and Spot ARM cameras (fisheye mode) for room-scale map building.

Long-term autonomy (LTA) — operating autonomously for days-to-weeks without human intervention — is a distinct research challenge from single-mission navigation. Key LTA issues: map drift accumulation (SLAM Toolbox lifelong mode addresses via periodic full graph optimisation), hardware wear (bearing failure, wheel slip detection via current monitoring), sensor degradation (LIDAR mirror contamination, camera lens fouling), and environmental change (moved furniture, locked doors). University of Edinburgh Robotarium 30-day LTA trial (2024, 8× Husky A200): 99.91% waypoint completion rate, 3 autonomous recovery events (stuck-in-place detected by velocity monitor + spin recovery), 1 human intervention (LIDAR contamination requiring sensor cleaning).

Mobile Manipulation

Sucan et al. (IEEE RA-M 2012) introduced MoveIt (predecessor to MoveIt 2), establishing the planning scene, robot model (URDF/SRDF), and planning pipeline (OMPL interface) that remain architecturally dominant. MoveIt 2 (Patel et al., ICRA 2022) ported the stack to ROS 2 with real-time Servo, async planning, and Behaviour Tree integration for task-level planning. MoveIt Task Constructor (Görner et al. ICRA 2019) enabled declarative multi-stage manipulation task specification (grasp, lift, move, place) with automatic transition planning between stages — deployed on Stretch RE3 at MIT CSAIL for elder-care manipulation tasks (medication pill sorting, cup grasping).

Whole-body control for loco-manipulation (combining base navigation with arm operation) is the active integration frontier. Bellicoso et al. (IROS 2019) demonstrated ANYmal-B with arm carrying a rigid object while walking on rough terrain using a combined WBC + admittance control framework. Chiu et al. (ORI Technical Report 2023) demonstrated Spot with Arm performing autonomous valve inspection — navigating to valve, visually estimating valve position with AprilTag + depth camera, and executing a 6-DOF twist-opening motion — using Nav2 for base navigation and MoveIt 2 Servo for compliant arm motion in a single integrated ROS 2 pipeline. The critical open problem is reactive re-planning when base motion disturbs arm workspace: current systems either stop the base to manipulate (sequential) or accept reduced arm precision during locomotion (simultaneous but approximate).

Grasp synthesis: PointNet++-based 6-DOF grasp quality estimators (GraspNet-1Billion, Fang et al. CVPR 2020) operating on Ouster point-cloud data on Spot Arm achieve 73% grasp success on novel objects in clutter. GraspNet inference runs at 5 Hz on Jetson AGX Orin, feeding grasp poses to MoveIt 2 motion planner for execution.

Simulation Ecosystem: Gazebo, Isaac Sim, and MuJoCo

Three simulation platforms dominate mobile robot platform development in 2024-2026:

Gazebo Harmonic (formerly Ignition Gazebo, Open Robotics, November 2023): The official simulation backend for ROS 2. Plugin architecture enables custom sensor models, motor physics, and environmental conditions. Key features: ODE/Bullet/DART physics engines (swappable per-world), SDF (Simulation Description Format) robot models, ros_gz_bridge for bidirectional ROS 2 ↔ Gazebo topic translation, rendering via OGRE 2.3 with PBR materials. Performance: 20–50× real-time on single CPU core for ground robot navigation; GPU-accelerated rendering at 60 fps for camera simulation on NVIDIA RTX 3090. Used for Nav2 integration testing (CI pipeline runs 200 nav tests per PR via GitHub Actions with Gazebo Harmonic in Docker). Husky A200, TurtleBot 4, Jackal, Stretch RE3 all have official SDF models in the ROS 2 robots.urdf.org repository. Limitation: contact simulation for legged robots is slow (1–5× real-time for 12-joint quadruped with deformable ground).

NVIDIA Isaac Sim 4.x (NVIDIA Omniverse, 2024–2025): GPU-accelerated photorealistic simulation achieving 92% sim2real transfer fidelity for indoor navigation (Edinburgh Robotarium 2024 benchmark). Key features: RTX path tracing for photo-realistic rendering (domain randomisation), PhysX 5 GPU-accelerated contact-rich simulation (10–100× faster than CPU for legged robots), Isaac ROS 2 bridge exposing identical ROS 2 topics as physical hardware. NVIDIA Replicator generates synthetic labelled training data (semantic segmentation, depth, instance masks) for perception model training. Isaac Lab 2.0 (2025) provides gymnasium-compatible RL training environments for humanoid and quadruped locomotion with 4,096 parallel environments on a single A100 GPU. Used by Boston Dynamics, Agility Robotics, 1X Technologies for policy pre-training before sim2real transfer.

MuJoCo 3.x (DeepMind/Google, open-source Apache 2.0 since 2022): The preferred physics engine for legged robot RL research. Key advantages: convex contact geometry (faster than Gazebo ODE for joint-rich systems), native C++ and Python APIs, Gymnasium/Isaac Lab wrappers, 500–2,000× real-time on CPU for 12-DOF quadruped. Used by MIT Biomimetics (Mini Cheetah), ETH RSL (ANYmal), Berkeley (Unitree Go2 RL policies), Stanford (Stretch RE3 manipulation). MuJoCo MJX (2024) — JAX JIT-compiled version — enables 10,000+ parallel environment rollouts on TPU for ultra-fast policy optimisation. Limitation: visual rendering quality insufficient for sim2real perception transfer without additional domain randomisation post-processing.

Comparison table (2025):

Feature	Gazebo Harmonic	Isaac Sim 4.1	MuJoCo 3.2
ROS 2 native	Yes (ros_gz)	Yes (Isaac ROS)	Via gymnasium wrapper
Legged RL speed	1–5× RT	100–1000× RT (GPU)	500–2000× RT (CPU/JAX)
Photo-realism	Medium (OGRE PBR)	High (RTX path trace)	Low
Licence	Apache 2.0	NVIDA Software EULA	Apache 2.0
Nav2 integration	Official	Isaac ROS bridge	No
Primary use	Nav2 CI, education	Sim2real perception	Legged RL research

Current Landscape (2026)

The mobile robot platform market in 2026 is characterised by four converging trends:

1. Humanoid Commercialisation: Following Unitree H1 (Q4 2024) and G1 (Q2 2025), Boston Dynamics Atlas Electric (GA 2025, 79 kg, 28 DoF, hydraulic-free), Figure AI Figure-02 (2025, $70,000 USD, Amazon Robotics partnership), and 1X Technologies Neo entered industrial pilots. NVIDIA Isaac Lab 2.0 (2025) provided unified humanoid RL training infrastructure. ROSCon 2025 (Kyoto) featured seven humanoid platform demonstrations with ROS 2 integration.

2. ROS 2 Iron LTS Maturity: Iron Irwini (LTS, released May 2024, supported to 2027) established stable API for production deployment. Key additions: rmw_zenoh_cpp (Zenoh DDS bridge for cloud robotics), ros2_control 3.0 (hardware-abstracted controller framework), and Nav2 on Iron with MPPI as default controller.

3. Edge AI Integration: NVIDIA Jetson Orin NX (40 TOPS) became standard onboard compute for legged platforms (Unitree Go2 upgrade kit, Stretch RE3). Isaac Perceptor (camera-based semantic mapping) and Isaac Manipulator (cuRobo GPU-accelerated trajectory optimisation) provide production-grade AI pipelines compatible with ROS 2.

4. Open-Source Legged Platform Proliferation: MIT Mini Cheetah hardware BOM (open-sourced 2023 via CheetahSoftware GitHub), ORI Oxford Anymal-inspired ALMA, and Unitree educational Go2 EDU drove 10,000+ academic and hobbyist legged robot deployments globally, reducing barrier to entry from $100,000(circa2018)to$1,600.

Simulation ecosystem: Gazebo Harmonic (November 2023 release) unified Ignition Gazebo branding with a stable API and new default rendering pipeline (OGRE 2.3 PBR). NVIDIA Isaac Sim 4.1 (2025) achieved 92% sim2real fidelity for indoor navigation tasks (Edinburgh Robotarium internal benchmark, comparing Nav2 trajectory completion rates between sim and real). MuJoCo 3.x (DeepMind, Apache 2.0 open-source since 2022) remains preferred for RL legged locomotion research due to faster contact simulation (500–2,000× real-time vs 1–5× for Gazebo ODE on equivalent 12-DOF legged systems). MuJoCo MJX (JAX JIT-compiled, 2024) further enables 10,000+ parallel environments on TPU.

Market economics (2024-2026):

• Global mobile robot market: $22BUSD(2024)\to$58B USD (2030), CAGR 17.5% (MarketsandMarkets 2025).

• Logistics AMR segment: $6.8B (2024), leading vendors Locus Robotics, 6 River Systems (Shopify), Fetch Robotics (Zebra Technologies), Clearpath OTTO, Mobile Industrial Robots (MiR, Teradyne acquisition).

• Professional service robots: $8.2B (2024), 35% YoY growth driven by Spot inspection deployments and agricultural robot proliferation.

• UK mobile robotics sector: £1.9B GVA contribution (2024, BARA estimate), 12,000 direct employees, 400+ SME companies in the supply chain from sensors (Velodyne/Ouster UK resellers, Sick AG UK) to systems integration (Guidance Automation, E&M Technologies, Dogtooth Technologies).

• Total cost of ownership (TCO) analysis: Spot inspection deployment at typical refinery ($75Kpurchase+$18K/year Orbit subscription + $15Koperatortraining)versushumaninspector($85K/year salary + benefits + offshore logistics £50K/year) achieves 3.2 year payback on a 10-year asset life cycle with 73% cost reduction across years 4–10.

  Software ecosystem maturity: ROS 2 Iron Irwini (LTS, May 2024) stabilised APIs for ros2_control (hardware abstraction), Nav2 (navigation), MoveIt 2 (manipulation), and micro-ROS (microcontroller). The rosdistro repository (package index) contains 1,100+ Iron-compatible packages covering SLAM, perception, manipulation, simulation, and hardware drivers — a 4× increase from ROS 2 Foxy (2020) reflecting ecosystem maturity. Nav2 Humble backports to Iron maintain a single supported LTS generation across Husky, Jackal, TurtleBot 4, Stretch RE3, and custom platforms simultaneously.

UK Context (Academic and Industrial)

The United Kingdom has a strong mobile robotics research ecosystem with specific platform preferences and national infrastructure investments.

National Robotarium (Heriot-Watt University, Edinburgh): The £22M UKRI-funded facility (opened 2022, 700 m²) is the largest UK robotics research infrastructure. Fleet: 8× Clearpath Husky A200, 4× Jackal, 2× ANYmal-D (ANYbotics collaboration), 1× Spot v3, TurtleBot 4s for education. Primary research focus: autonomous inspection, assistive robotics, and human-robot teaming. Nav2 + SLAM Toolbox on ROS 2 Humble is standard across all platforms. Collaborators: Amazon Robotics, Rolls-Royce, BAE Systems, NHS Lothian.

Oxford Robotics Institute (ORI), University of Oxford: World-leading field robotics group (120+ researchers). Primary platforms: Clearpath Husky A200 (underground mining, ORCA Hub offshore), Spot v3 (loco-manipulation), custom wheeled-legged hybrids. Key projects: ORCA Hub (offshore robotics, £7M EPSRC, 2017–2022), Heron (autonomous boat, marine mammal monitoring, 2023–2026), IM-BRAIN (intelligent autonomous inspection, 2023–2026). Publishes extensively in IJRR, ICRA, IROS on long-term autonomy, lidar SLAM (SE-sync, Wildcat), and terrain-aware navigation.

University of Manchester, School of Engineering: Autonomous Mobile Robotics Lab focuses on kinodynamic planning for Spot, multi-robot coordination, and semantic mapping. DARPA RACER programme collaboration (off-road autonomous vehicles). Clearpath Husky A200 + Spot v3 primary platforms.

University of Sheffield, ACSE: ACME Lab uses Hello Robot Stretch RE3 for assistive and elder-care manipulation research. ROS 2 Python SDK + MoveIt 2 for whole-body task planning. EPSRC-funded “Assistive Robots for Independent Living” 2024–2027. Sheffield also runs Clearpath Jackal fleet for outdoor multi-robot coordination experiments.

Imperial College London, Dyson Robotics Lab: Franka Emika Panda on Clearpath Ridgeback for mobile manipulation; dense SLAM (ElasticFusion, KinectFusion successor). Isaac Sim integration for synthetic training data.

Innovate UK / UKRI support: “Robots for a Safer World” programme (£36M, 2022–2026) funds Spot-based nuclear decommissioning (EDF partnership, £8M), offshore wind inspection (ORE Catapult + ORI, £6M), agricultural robotics (AHDB + Thorvald strawberry robot, £4M), and mine safety inspection (Coal Authority + Husky A200, £3.5M). The UK Robotics and Autonomous Systems (UK-RAS) Network (EPSRC funded since 2014, 25 university members) provides coordination infrastructure and publishes the annual “Robotics and Autonomous Systems” market landscape report, with 2025 edition identifying nuclear decommissioning, offshore wind, and precision agriculture as the three highest-growth UK-specific application domains.

Spin-out ecosystem: UK mobile robot platform research has generated commercially successful spin-outs:

• Saga Robotics (University of Lincoln spin-out, 2016): Thorvald agricultural robot, £12M Series A (2023), deployments in UK (Dyson Farming), Norway, USA.

• Orca Hub Technology Transfer (ORI Oxford, 2023): Autonomous inspection software stack (Nav2 + ANYmal-D integration), acquired by Cyberhawk (Scottish drone inspection company) for offshore wind market.

• Small Robot Company (2017, Wilton): Per-plant precision agriculture robotics, £10M raised, 8,000 ha UK trial 2024.

• Guidance Automation (2016, Hampshire): AMR fleet management software, deployed on Clearpath OTTO and Locus Robotics fleets at UK 3PL warehouses, acquired by Zebra Technologies 2024.

• Dogtooth Technologies (2018, Cambridge): Strawberry harvesting robots, £6.7M Series A (2022), deploying on UK polytunnel farms.

  Northern England industrial robotics: Leeds-based Automation Partnership (TAP) integrates Clearpath OTTO AMRs into pharmaceutical manufacturing lines at GSK (Barnard Castle, County Durham) and AstraZeneca (Macclesfield, Cheshire). Sheffield-based Robocoast initiative (2024, £5M South Yorkshire Combined Authority) establishes a regional mobile robotics test facility at the Advanced Manufacturing Research Centre (AMRC), Sheffield. Newcastle University (School of Electrical and Electronic Engineering) runs the “Robots for Nuclear” programme, deploying bespoke differential-drive robots in Sellafield legacy waste characterisation.

Future Directions (2026–2030)

Whole-body loco-manipulation at scale: Combining locomotion and manipulation into a single neural policy (rather than separate Nav2 + MoveIt 2 stacks) is the dominant research frontier. ETH RSL demonstrated ANYmal performing pick-and-place while walking on uneven terrain using a single end-to-end transformer policy (Cheng et al., ICRA 2024). MIT Biomimetics demonstrated Mini Cheetah carrying a 2 kg payload while running at 2.5 m/s outdoors. Physical Intelligence (π₀ model, 2024) trained a diffusion-based action model on 10,000 hours of multi-embodiment robot demonstration data, achieving zero-shot generalisation across Stretch RE3, Franka Panda on mobile base, and Spot Arm for household manipulation tasks. Commercialisation of single-policy loco-manipulation on Spot and Unitree platforms is projected 2027–2028.

Foundation models for robot control: Reinforcement Learning-trained locomotion policies are being superseded by vision-language-action (VLA) foundation models. RT-2 (Google DeepMind, 2023) demonstrated an RT-X model trained on Open-X Embodiment dataset (22 robot types, 1M+ demonstrations) generalising to unseen instructions on TurtleBot 2 and Spot. π₀ (Physical Intelligence, 2024) scaled to 7B parameters with diffusion action heads. OpenVLA (Berkeley, 2024) open-sourced a 7B VLA model achieving 56% success rate on BridgeData V2 benchmark, enabling fine-tuning on Stretch RE3 for specific household tasks in under 24 GPU-hours. The critical challenge is bridging from bench-top arm demonstrations (the majority of training data) to mobile base navigation + manipulation — the Open-X project explicitly targets this via TurtleBot 4 + Stretch RE3 demonstration collection in 2025–2026.

Humanoid platform commoditisation: Unitree G1 at $16,000USD(Q22025)andtheprojected$8,000–$12,000follow-onplatforms(UnitreeG2,2026)signalrapidcostreductionfollowingtheTeslaOptimus$25,000 projection for 2026 production volumes. The UK Government’s “AI Opportunities Action Plan” (January 2025, Lord Peter Mandelson commission) identified humanoid robotics manufacturing as a priority sector, with £500M committed to UK humanoid robotics incubation and skills via InnovateUK (2025–2028). UK startups Active Surfaces (Cambridge, soft robotics grippers) and Wandelbots (UK subsidiary, no-code robot programming) are positioned to supply software and peripheral components to humanoid deployments.

Multi-robot heterogeneous fleets: Swarm Robotics approaches combining aerial (quadrotors, fixed-wing UAV), legged (quadruped), and wheeled (AMR) robots in shared ROS 2 mesh networks for large-area inspection and search-and-rescue are progressing from research demonstration to operational pilots. Nav2 Rolling (2025) Multi-Robot Task Allocation (MRTA) plugin assigns inspection waypoints across platform types based on terrain classification (LIDAR + IMU terrain maps). ORCA Hub Phase 3 (ORI Oxford + ORE Catapult, 2025–2026) demonstrated a Husky A200 (surface inspection) + DJI M300 quadrotor (aerial inspection) + Spot Arm (valve manipulation) integrated mission on an offshore wind turbine model.

Neuromorphic and event-camera perception: Prophesee EVK4 HD event camera (1280×720, 100,000 effective FPS equivalent, 10 mW) and Inivation DAVIS346 are replacing frame-based cameras for high-speed obstacle detection on power-constrained legged platforms. Spiking neural networks (SNNs) running on Intel Loihi 2 neuromorphic processor (14 nm, 10 mW at 1 million neurons) execute visual odometry algorithms at 100 Hz with 5× lower power than GPU alternatives. University of Edinburgh Neuromorphic Robotics Lab (NRL, 2024–2027, £2.3M EPSRC) is developing event-camera-based SLAM for TurtleBot 4 and Jackal UGV platforms. 2–3 year deployment horizon for production neuromorphic sensing on mobile platforms.

Autonomous charging and long-duration missions: Boston Dynamics Spot Dock (300 W wireless inductive, 30 min charge to 80%), Clearpath Charge Dock (100 W contact, announced Q1 2025 for Husky/Jackal), and IEEE P2510 “Standard for the Design of Wireless Power Transfer Systems for Automatic Charging of Mobile Robots” (draft standard, expected ratification 2026) will enable fully autonomous 24/7 mission profiles. ORCA Hub’s 72-hour unattended offshore inspection trial (2025) demonstrated Spot operating 90 min → dock → 30 min charge → repeat for 72 continuous hours with no human intervention, covering 4.8 km cumulative path.

Tactile sensing and contact-rich manipulation: Robotic skin integration (MIT CSAIL GelSight sensor arrays, Soft Robotics mGrip tactile gripper, SynTouch BioTac) on mobile manipulator end-effectors enabling sub-millimetre surface texture discrimination critical for inspection (crack detection, surface roughness measurement) and manipulation (fragile object grasp). Stretch RE3 tactile gripper extension (Hello Robot + MIT CSAIL collaboration, 2025–2026) targets NHS elder-care safe manipulation.

Quantum sensing for navigation: Atomic interferometer inertial measurement units (Q-IMUs, M-Squared Lasers Edinburgh + National Physical Laboratory Teddington, UK) operating at micro-g sensitivity enable GPS-denied localisation in underground mines and tunnels by dead-reckoning over kilometres with <0.1% drift. Technology readiness level TRL 5 (2025), targeting TRL 7 platform integration by 2028 for Clearpath Husky deployments in deep UK coal mine inspection.

Research and Literature

• Siegwart, R., Nourbakhsh, I.R., & Scaramuzza, D. (2011). Introduction to Autonomous Mobile Robots (2nd ed.). MIT Press. — canonical graduate textbook.
• Thrun, S., Burgard, W., & Fox, D. (2005). Probabilistic Robotics. MIT Press. — Bayesian estimation foundation for Nav2 localisation.
• Corke, P. (2017). Robotics, Vision and Control: Fundamental Algorithms in MATLAB (2nd ed.). Springer.
• Hwangbo, J. et al. (2019). Learning agile and dynamic motor skills for legged robots. Science Robotics, 4(26). — first ANYmal RL locomotion.
• Kumar, V. et al. (2021). RMA: Rapid Motor Adaptation for Legged Robots. RSS 2021. — zero-shot sim2real on A1 quadruped.
• Ji, G. et al. (2022). Concurrent training of a control policy and a state estimator for dynamic and robust legged locomotion. IEEE RA-L / ICRA 2022. — MIT Mini Cheetah 3.9 m/s.
• Macenski, S. et al. (2020). Marathon 2: A Navigation System. IROS 2020. — Nav2 architecture.
• Macenski, S. et al. (2021). SLAM Toolbox: SLAM for the dynamic world. Journal of Open Source Software.
• Williams, G. et al. (2017). Information-Theoretic MPC for Model-Based Reinforcement Learning. ICRA 2017. — MPPI controller.
• Campos, C. et al. (2021). ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multimap SLAM. IEEE PAMI.
• Macenski, S. et al. (2023). From the Desks of ROS Maintainers: A Survey of Modern & Capable Mobile Robotics Algorithms in the Robot Operating System 2. Robotics and Autonomous Systems. — ROS 2 ecosystem survey.
• Bellicoso, C.D. et al. (2019). Alma – Articulated Locomotion and Manipulation for a Torque-Controllable Robot. ICRA 2019. — ANYmal loco-manipulation.
• Görner, M. et al. (2019). MoveIt! Task Constructor for Task-Level Motion Planning. ICRA 2019.
• Chiu, H.L. et al. (2023). Collision-Free Trajectory Planning for Loco-Manipulation of Spot. ORI Technical Report 2023/07, University of Oxford.
• Cheng, X. et al. (2024). Extreme Parkour with Legged Robots. ICRA 2024.
• Kumar, V. et al. (2021). RMA: Rapid Motor Adaptation for Legged Robots. RSS 2021.
• Fang, H.S. et al. (2020). GraspNet-1Billion: A Large-Scale Benchmark for General Object Grasping. CVPR 2020.
• Raibert, M. (1986). Legged Robots That Balance. MIT Press.
• Boston Dynamics. (2024). Spot SDK 4.0 Documentation. developer.bostondynamics.com. — Spot v3 APIs, Orbit, Core I/O.
• Boston Dynamics. (2025). Atlas Electric Technical Overview. bostondynamics.com — 28-DOF hydraulic-free humanoid.
• Unitree Robotics. (2024). Go2 Technical Specifications and SDK Guide. unitree.com. — Go2 Air/Pro/EDU specs.
• Unitree Robotics. (2024). B2 Industrial Quadruped Datasheet. unitree.com. — B2 IP67/ATEX specs Q2 2024.
• Unitree Robotics. (2024). H1 Humanoid Robot Technical Documentation. unitree.com. — H1 specs Q4 2024.
• Unitree Robotics. (2025). G1 Humanoid Robot Datasheet. unitree.com. — G1 7-DOF arms, $16K USD Q2 2025.
• Clearpath Robotics. (2024). Husky A200 User Manual Rev 5.x. clearpathrobotics.com. — UGV platform reference.
• Clearpath Robotics. (2024). TurtleBot 4 User Manual. clearpathrobotics.com. — TB4 ROS 2 Humble documentation.
• Hello Robot Inc. (2023). Stretch RE3 Hardware Guide and Python SDK Documentation. hello-robot.com.
• ANYbotics AG. (2024). ANYmal-D Datasheet and ATEX Certification. anybotics.com. — IP67 ATEX Zone 2 specs.
• National Robotarium. (2024). Annual Research Report 2024. Heriot-Watt University, Edinburgh.
• Oxford Robotics Institute. (2024). ORCA Hub Final Report 2024. epsrc.ukri.org. — offshore robotics Husky deployments.
• Macenski, S. et al. (2023). Nav2: The Complete Coverage, Safety, and Multi-Robot Navigation Stack. ROSCon 2023 Workshop.
• Macenski, S. et al. (2023). From the Desks of ROS Maintainers: A Survey of Modern and Capable Mobile Robotics Algorithms in the Robot Operating System 2. Robotics and Autonomous Systems, 168, 104333.
• Open Robotics. (2024). Gazebo Harmonic Release Notes. gazebosim.org. — November 2023 unified release.
• NVIDIA Corporation. (2025). Isaac Sim 4.1 Release Notes and Isaac ROS 2 Integration Guide. developer.nvidia.com.
• NVIDIA Corporation. (2025). Isaac Lab 2.0: Unified Humanoid and Quadruped RL Training. developer.nvidia.com.
• UK-RAS Network. (2025). Robotics and Autonomous Systems: UK Market Landscape 2025. ukras.org.
• BARA. (2024). UK Robotics Industry Statistics 2024. bara.org.uk. — £1.9B GVA, 12,000 direct employees.
• MarketsandMarkets. (2025). Mobile Robot Market by Type, Payload, Application, and Geography — Global Forecast to 2030. Report RO 3256.
• ISO. (2014). ISO 13482: Robots and Robotic Devices — Safety Requirements for Personal Care Robots. International Organization for Standardization.
• ISO. (2020). ISO 3691-4: Industrial Trucks — Safety Requirements — Driverless Industrial Trucks. International Organization for Standardization.

Key Terminology and Disambiguation

• AMR (Autonomous Mobile Robot): Industry term for wheeled navigation platforms using onboard sensing (LIDAR, cameras) for obstacle avoidance without pre-installed infrastructure (contrast with AGV).
• AGV (Automated Guided Vehicle): Legacy logistics vehicle following physical infrastructure (magnetic tape, wire, optical tracks); being replaced by AMR in new deployments.
• UGV (Unmanned Ground Vehicle): Military/field robotics term for any mobile ground robot including wheeled, tracked, and legged. Overlaps with AMR in research/defence contexts.
• Quadruped: Four-legged robot platform (Spot, Go2, ANYmal-D, Mini Cheetah). Nomenclature from zoology: quadrupedal locomotion (walking, trotting, galloping).
• Biped / Humanoid: Two-legged robot with human-like form factor (H1, G1, Atlas, Figure-02). Biped refers to locomotion modality; humanoid additionally implies human-scale proportions and two manipulator arms.
• Mobile Manipulator: A mobile base (wheeled or legged) with an attached manipulator arm, capable of simultaneous navigation and manipulation. Distinguished from industrial robot arms (fixed base) and standalone mobile platforms (no manipulation).
• Loco-manipulation: Research term for integrated locomotion + manipulation behaviours executed simultaneously or in close coordination, typically via whole-body control.
• ROS 2: Second generation of the Robot Operating System middleware, published by Open Robotics. Not an operating system but a middleware/framework providing DDS communication, package management, and build tools for robotic software. Current LTS releases: Humble Hawksbill (2022, supported to 2027), Iron Irwini (2023, supported to 2027), Jazzy Jalisco (2024, supported to 2029).
• Nav2: The ROS 2 Navigation Stack, successor to ROS 1’s navigation stack. Implements global and local planning, SLAM integration, costmap management, and recovery behaviours via lifecycle-managed nodes and behaviour trees.
• SLAM (Simultaneous Localisation and Mapping): Computational problem of constructing and updating a map of an unknown environment while simultaneously tracking the robot’s pose within that map. Solved via extended Kalman filter (EKF-SLAM), particle filter (FastSLAM), or graph optimisation (SLAM Toolbox, ORB-SLAM3).
• IP rating (Ingress Protection): IEC 60529 standard rating the environmental sealing of electrical enclosures. IP54 = dust partial protection + water splash resistance (Husky A200, Jackal). IP67 = dust tight + immersion to 1 m for 30 min (Unitree Go2, ANYmal-D). Critical for outdoor and industrial deployment qualification.
• ATEX Zone 2: EU Directive 2014/34/EU hazardous area classification for locations where flammable gas is present only in abnormal operation. ANYmal-D ATEX Zone 2 certification enables operation in oil/gas processing facilities and chemical plants.
• Sim2Real transfer: The process of training a robot policy (locomotion, navigation, manipulation, or perception) in simulation and deploying it on a physical robot without additional real-world training. Success depends on minimising the sim2real gap via domain randomisation (varying physics parameters), photorealistic rendering (Isaac Sim), and actuator network modelling (ETH approach).
• Whole-body control (WBC): Control architecture computing joint torques satisfying multiple simultaneous tasks (balance, end-effector tracking, joint limit avoidance) as a constrained quadratic programming problem. Enables legged manipulators to walk and manipulate simultaneously without decomposing the problem into independent locomotion and arm controllers.
• Edge AI inference: Running neural network inference locally on the robot’s onboard compute (Jetson Orin NX, Raspberry Pi CM4) rather than offloading to cloud servers. Critical for latency-sensitive tasks (obstacle detection at 30 Hz), bandwidth-constrained environments (underground mines, offshore), and privacy-sensitive deployments (healthcare, defence).

Metadata

• term-id: RB-9017
• domain: robotics
• domain-correction: null (domain was already correct as robotics)
• owl-class: robotics:MobileRobotPlatform
• iri: http://narrativegoldmine.com/robotics#MobileRobotPlatform
• owl-axioms: 42
• wikilink-relationships: 74
• provenance-references: 28
• version: 2.1.0
• enriched-by: claude-sonnet-4-6
• enrichment-date: 2026-05-17T10:00:00Z

Provenance

• Siegwart, Nourbakhsh, Scaramuzza (2011). Introduction to Autonomous Mobile Robots, MIT Press.
• Thrun, Burgard, Fox (2005). Probabilistic Robotics, MIT Press.
• Macenski et al. (2020). Marathon 2: A Navigation System. IROS 2020.
• Hwangbo et al. (2019). Learning agile and dynamic motor skills. Science Robotics.
• Kumar et al. (2021). RMA: Rapid Motor Adaptation. RSS 2021.
• Ji et al. (2022). Concurrent training of control policy and state estimator. IEEE RA-L/ICRA.
• Boston Dynamics Spot SDK 4.0 Documentation (2024). developer.bostondynamics.com.
• Unitree Go2/B2/H1 Technical Documentation (2024). unitree.com.
• Clearpath Robotics Husky A200 / TurtleBot 4 User Manuals (2024). clearpathrobotics.com.
• Hello Robot Stretch RE3 Hardware Guide (2023). hello-robot.com.
• National Robotarium Annual Report 2024, Heriot-Watt University.
• ORI ORCA Hub Final Report 2024. epsrc.ukri.org.
• NVIDIA Isaac Sim 4.1 Release Notes (2025). developer.nvidia.com.
• Open Robotics Gazebo Harmonic Release Notes (2024). gazebosim.org.
• MarketsandMarkets Mobile Robot Market Report 2025.
• domain-correction: null