Title: Entropic-Regularized Second-Order Dynamic OptimizationDate: June 26, 2026Time: 12:00 PMLocation: Coda Conference Room C1115 Yuichiro AoyamaMachine Learning PhD CandidateSchool of Aerospace EngineeringGeorgia Institute of Technology CommitteeDr. Evangelos A. Theodorou (Advisor), School of Aerospace Engineering, Georgia TechDr. Glen Chou, School of Aerospace Engineering, Georgia Tech Dr. Kyriakos G. Vamvoudakis, School of Aerospace Engineering, Georgia TechDr. Frank Dellaert, School of Interactive Computing, Georgia TechDr. Kenshiro Oguri, School of Aeronautics and Astronautics, Purdue University   AbstractThis dissertation addresses the challenge of performing optimal control for dynamical systems operating in non-convex cost landscapes arising from nonlinear dynamics and cluttered environments. While classical second-order methods like Differential Dynamic Programming (DDP) offer fast local convergence, they easily get trapped in local minima due to their reliance on local information. Although purely sampling-based methods bypass the locality issue, they suffer from high sample complexity and noisy decision-making. To bridge this gap, this work leverages the Maximum Entropy DDP (ME-DDP) framework, in which a structured exploration covariance naturally arises from entropic regularization, balancing second-order local exploitation with robust trajectory perturbation. We extend this mechanism beyond standard Shannon entropy to generalized representations, detailing the development of Tsallis entropy and Stein Variational DDP (SV-DDP) to maintain policy diversity without sacrificing optimization structure. Benchmarked against Model Predictive Path Integral (MPPI) variants within a Model Predictive Control (MPC) framework, the proposed algorithms demonstrate superior overall performance, with their practical robustness validated through hardware experiments on a quadrotor navigation task in cluttered environments.