报告题目：Learning-Based Control With Planning in Adversarial Environments

报告人：Carmen Del Vecchio，University of Sannio

报告时间：2026年11月18日（周三）9:30-11:30

报告地点：20-404

报告摘要：Learning-based control plans with a model learned from data, and is far more sample-efficient than model-free reinforcement learning. The learned model is never exact, however, and planners that quantify its uncertainty still optimize the average return, with no guarantee against systematic or adversarial mismatch. We present a Distributionally Robust (DR) framework for learning-based planning: the learned model is the centre of a Wasserstein ball over models, and planning targets the worst case in the ball. Wasserstein duality turns the intractable max-min problem into a convex one with a regularized cost, so robustness becomes regularization. The framework is shown in the context of PETS (Probabilistic Ensamble with Trajectory Planning) algorithm, yielding DR-PETS, with improved robustness to parameter shifts on benchmark tasks.

报告人简介：Carmen Del Vecchio received the M.Sc. degree (five years) in management engineering from the University of Naples Federico II, Naples, Italy, in 1998, and the Ph.D. degree in industrial automation from the University of Sannio, Benevento, Italy, in 2003. She is an Associate Professor in automatic control with the Department of Engineering, University of Sannio, Benevento, Italy. Her research interests include the optimization and control of interconnected processes through various theoretical tools (e.g., decision and Markov chains, probabilistic Boolean networks, machine learning, and reinforcement learning) and their applications in several fields including manufacturing, smart grids, medicine, and biology. Dr. Del Vecchio is currently an Associate Editor for Automatica (Elsevier), IEEE Control Systems Letters, Nonlinear Analysis: Hybrid System (IFAC), and International Journal of Robust and Nonlinear Control (Wiley). She is also a Member of the Conference Editorial Board of the IEEE Control Systems Society. Since November 2025, she has been the Chair of the bachelor’s degree program in electrical engineering and Biomedicine and of the master’s degree program in electronic engineering for automation and sensing with the University of Sannio.

邀请人：刘洋