Safe Exploration in Markov Decision Processes

Teodor Mihai Moldovan, Pieter Abbeel

In environments with uncertain dynamics exploration is necessary to learn how to perform well. Existing reinforcement learning algorithms provide strong exploration guarantees, but…