A mathematical formulation in reinforcement learning that maps state-action pairs to scalar values, guiding AI agent behaviour toward desired outcomes through feedback signals; central to policy optimisation, agent training, and objective specification in machine learning systems.
Semantic Classification
Content
Key Concepts
- Sparse vs dense reward signals
- Reward shaping for faster convergence
- Inverse reinforcement learning
- Multi-objective reward functions
- Credit assignment problem
Applications
- Game NPC behaviour optimization
- Virtual character training
- Autonomous vehicle navigation
- Robotic control systems
- Player engagement optimization