Efficient Solution and Learning of Robust Factored MDPs
This work addresses the challenge of synthesizing robust policies with provable guarantees in uncertain environments, offering incremental improvements in sample efficiency for domain-specific applications like robotics or control systems.
The paper tackles the problem of learning robust Markov decision processes (r-MDPs) with factored state-space representations to improve sample efficiency, showing that their methods yield dimensional gains and produce more effective robust policies with tighter performance guarantees than state-of-the-art methods.
Robust Markov decision processes (r-MDPs) extend MDPs by explicitly modelling epistemic uncertainty about transition dynamics. Learning r-MDPs from interactions with an unknown environment enables the synthesis of robust policies with provable (PAC) guarantees on performance, but this can require a large number of sample interactions. We propose novel methods for solving and learning r-MDPs based on factored state-space representations that leverage the independence between model uncertainty across system components. Although policy synthesis for factored r-MDPs leads to hard, non-convex optimisation problems, we show how to reformulate these into tractable linear programs. Building on these, we also propose methods to learn factored model representations directly. Our experimental results show that exploiting factored structure can yield dimensional gains in sample efficiency, producing more effective robust policies with tighter performance guarantees than state-of-the-art methods.