ROSARL: Reward-Only Safe Reinforcement Learning
This addresses the challenge of safe reinforcement learning for agents in complex environments, offering a practical solution to avoid unsafe states without manual penalty tuning.
The paper tackles the problem of designing safe reinforcement learning agents by introducing the 'Minmax penalty', an upper bound on rewards at unsafe states that minimizes the probability of reaching those states, and demonstrates its effectiveness in high-dimensional continuous control environments.
An important problem in reinforcement learning is designing agents that learn to solve tasks safely in an environment. A common solution is for a human expert to define either a penalty in the reward function or a cost to be minimised when reaching unsafe states. However, this is non-trivial, since too small a penalty may lead to agents that reach unsafe states, while too large a penalty increases the time to convergence. Additionally, the difficulty in designing reward or cost functions can increase with the complexity of the problem. Hence, for a given environment with a given set of unsafe states, we are interested in finding the upper bound of rewards at unsafe states whose optimal policies minimise the probability of reaching those unsafe states, irrespective of task rewards. We refer to this exact upper bound as the "Minmax penalty", and show that it can be obtained by taking into account both the controllability and diameter of an environment. We provide a simple practical model-free algorithm for an agent to learn this Minmax penalty while learning the task policy, and demonstrate that using it leads to agents that learn safe policies in high-dimensional continuous control environments.