A Glossary of Robot Control and AI Terms

The vocabulary that comes up constantly in any discussion of how modern robots actually move and decide.

Control basics

Policy: the function mapping sensor input to robot action. Controller: the software (classical or learned) that turns a desired behavior into actual joint commands. Feedback loop: a control system that continuously adjusts its output based on measured results, rather than executing a fixed, open-loop plan.

Learning-specific terms

Reward function: what a reinforcement learning policy is trained to maximize. Domain randomization: varying simulated conditions during training to improve real-world robustness. Sim-to-real gap: the performance difference between a policy's simulated and real-world behavior. Behavior cloning: training a policy to imitate recorded demonstrations directly, without a reward signal.