Abstract: Current constrained reinforcement learning (RL) methods guarantee constraint satisfaction only in expectation, which is inadequate for safety-critical decision problems. Since a constraint ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results