<ul><li>Real-world decision-making problems often have complex, uncertain dynamics.</li><li>Traditional model-based reinforcement learning approaches don't consider underlying causal mechanisms, leading to spurious correlations.</li><li>Causal Model-Based Policy Optimization (C-MBPO) integrates causal learning into the MBRL pipeline.</li><li>C-MBPO infers a Causal Markov Decision Process (C-MDP) and learned Structural Causal Models (SCMs) for more robust and generalizable policy learning.</li></ul>

Towards Causal Model-Based Policy Optimization

Discover more