We present an analytical policy update rule that is independent of parameterized function approximators. The update rule is suitable for general stochastic policies with monotonic improvement guarantee. The update rule is derived from a closed-form trust-region solution using calculus of variation, following a new theoretical result that tightens existing bounds for policy search using trust-region methods. An explanation building a connection between the policy update rule and value-function methods is provided. Based on a recursive form of the update rule, an off-policy algorithm is derived naturally, and the monotonic improvement guarantee remains. Furthermore, the update rule extends immediately to multi-agent systems when updates are performed by one agent at a time.
We extend trust region policy optimization (TRPO) to multi-agent reinforcement learning (MARL) problems. We show that the policy update of TRPO can be transformed into a distributed consensus optimization problem for multi-agent cases. By making a series of approximations to the consensus optimization model, we propose a decentralized MARL algorithm, which we call multi-agent TRPO (MATRPO). This algorithm can optimize distributed policies based on local observations and private rewards. The agents do not need to know observations, rewards, policies or value/action-value functions of other agents. The agents only share a likelihood ratio with their neighbors during the training process. The algorithm is fully decentralized and privacy-preserving. Our experiments on two cooperative games demonstrate its robust performance on complicated MARL tasks.