Safety Gym
A safe reinforcement learning benchmark environment introduced by OpenAI for evaluating the ability of agents to complete tasks while satisfying safety constraints.
Timeline (last 90 days)
In the double integrator example and the Safety Gym environments, FAITH achieves the highest return among methods with no feasible-start violations, while matching the lowest harm level under infeasible starts.
Safety Gym is a safe reinforcement learning benchmark suite proposed by OpenAI, containing agents such as Point, Car, and Ant, with hazard regions and goals in the tasks, and using feasible/infeasible starts to distinguish whether a policy is within the safety boundary.
All Facts (2)
In the double integrator example and the Safety Gym environments, FAITH achieves the highest return among methods with no feasible-start violations, while matching the lowest harm level under infeasible starts.
50%UnverifiedSafety Gym is a safe reinforcement learning benchmark suite proposed by OpenAI, containing agents such as Point, Car, and Ant, with hazard regions and goals in the tasks, and using feasible/infeasible starts to distinguish whether a policy is within the safety boundary.
50%