Reinforcement Learning of Risk-Constrained Policies in Markov Decision Processes
| Authors | |
|---|---|
| Year of publication | 2020 |
| Type | Article in Proceedings |
| Conference | The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020 |
| MU Faculty or unit | |
| Citation | |
| Web | https://aaai.org/ojs/index.php/AAAI/article/view/6531 |
| Doi | http://dx.doi.org/10.1609/aaai.v34i06.6531 |
| Keywords | reinforcement learning; Markov decision processes; Monte Carlo tree search; risk aversion |
| Description | Markov decision processes (MDPs) are the defacto framework for sequential decision making in the presence of stochastic uncertainty. A classical optimization criterion for MDPs is to maximize the expected discounted-sum payoff, which ignores low probability catastrophic events with highly negative impact on the system. On the other hand, risk-averse policies require the probability of undesirable events to be below a given threshold, but they do not account for optimization of the expected payoff. We consider MDPs with discounted-sum payoff with failure states which represent catastrophic outcomes. The objective of risk-constrained planning is to maximize the expected discounted-sum payoff among risk-averse policies that ensure the probability to encounter a failure state is below a desired threshold. Our main contribution is an efficient risk-constrained planning algorithm that combines UCT-like search with a predictor learned through interaction with the MDP (in the style of AlphaZero) and with a risk-constrained action selection via linear programming. We demonstrate the effectiveness of our approach with experiments on classical MDPs from the literature, including benchmarks with an order of 10^6 states. |
| Related projects: |