Distributionally Robust MPPI (Model Predictive Path Integral) for Safe Decision Making

The second seminar in the 2026 CSIE Seminar Series took place, focusing on an in-depth exploration of Distributionally Robust Model Predictive Path Integral (DR-MPPI) control, covering its theoretical foundations, risk-aware safety constraints using CVaR, and applications to robot navigation in uncertain environments with moving obstacles.

The seminar was led by CSIE Junior Researchers Lusine Barseghyan and Anna Beklaryan, who presented comprehensive insights into the topic, discussing key concepts including Wasserstein ambiguity sets, Lipschitz regularization for DR-CVaR, optimal control distribution, and practical applications such as safe UAV navigation and robotic manipulation under uncertainty.