A Theoretical Justification for Asymmetric Actor-Critic Algorithms
This provides a theoretical foundation for a specific class of reinforcement learning algorithms, but it is incremental as it adapts existing analysis to a particular setting.
The paper tackles the lack of theoretical justification for asymmetric actor-critic algorithms in partially observable environments, showing that the asymmetric critic eliminates error terms from state aliasing in a finite-time convergence analysis.
In reinforcement learning for partially observable environments, many successful algorithms have been developed within the asymmetric learning paradigm. This paradigm leverages additional state information available at training time for faster learning. Although the proposed learning objectives are usually theoretically sound, these methods still lack a precise theoretical justification for their potential benefits. We propose such a justification for asymmetric actor-critic algorithms with linear function approximators by adapting a finite-time convergence analysis to this setting. The resulting finite-time bound reveals that the asymmetric critic eliminates error terms arising from aliasing in the agent state.