A robot policy determines how a robot acts from its current sensor information, internal state and task goal. It may be programmed with explicit rules, learned from demonstrations or optimized through reinforcement learning. The policy can produce high-level plans or direct motor commands.
A learned policy can adapt to varied situations but may behave unpredictably outside its training experience. Physical deployment therefore requires representative testing, constrained actions, monitoring and independent safety controls that do not rely only on the policy's judgment.
Acronyms and aliases
robot control policy variant
Related terms
Frequently asked questions
How is a robot policy learned?
A policy can be learned from human demonstrations, simulation, reinforcement learning, real-world data or a combination of these methods.
Why does a robot policy need safety controls?
A learned policy may encounter unfamiliar conditions or select an unsafe action, so external limits and monitoring are needed to protect people and equipment.
