Provably Efficient Model-Free Constrained RL with Linear Function ApproximationDownload PDF

Published: 31 Oct 2022, Last Modified: 05 Jan 2023NeurIPS 2022 AcceptReaders: Everyone
Keywords: Reinforcement Learning Theory, Theory of Constrained Reinforcement Learning, Linear MDP, Model-free RL, Soft-max
TL;DR: We provide the first $\tilde{\mathcal{O}}(\sqrt{T})$ regret and constraint violation for model-free RL algorithm for the constrained linear MDP. In the process, we improve the bounds on the tabular set-up as well.
Abstract: We study the constrained reinforcement learning problem, in which an agent aims to maximize the expected cumulative reward subject to a constraint on the expected total value of a utility function. In contrast to existing model-based approaches or model-free methods accompanied with a `simulator’, we aim to develop the first \emph{model-free}, \emph{simulator-free} algorithm that achieves a sublinear regret and a sublinear constraint violation even in \emph{large-scale} systems. To this end, we consider the episodic constrained Markov decision processes with linear function approximation, where the transition dynamics and the reward function can be represented as a linear function of some known feature mapping. We show that $\tilde{\mathcal{O}}(\sqrt{d^3H^3T})$ regret and $\tilde{\mathcal{O}}(\sqrt{d^3H^3T})$ constraint violation bounds can be achieved, where $d$ is the dimension of the feature mapping, $H$ is the length of the episode, and $T$ is the total number of steps. Our bounds are attained without explicitly estimating the unknown transition model or requiring a simulator, and they depend on the state space only through the dimension of the feature mapping. Hence our bounds hold even when the number of states goes to infinity. Our main results are achieved via novel adaptations of the standard LSVI-UCB algorithms. In particular, we first introduce primal-dual optimization into the LSVI-UCB algorithm to balance between regret and constraint violation. More importantly, we replace the standard greedy selection with respect to the state-action function with a soft-max policy. This turns out to be key in establishing uniform concentration (a critical step for provably efficient model-free exploration) for the constrained case via its approximation-smoothness trade-off. Finally, we also show that one can achieve an even zero constraint violation for large enough $T$ by trading the regret a little bit but still maintaining the same order with respect to $T$.
Supplementary Material: pdf
14 Replies

Loading