Speaker：Lei Wang (Dalian University of Technology)

Time：2026-5-23 16:00

Location：Conference Room C503 at Administration Building at Haiyun Campus

Abstract:

This presentation introduces advanced trajectory planning strategies for autonomous maritime and port systems, transitioning from robust mathematical optimization to real-time hierarchical learning frameworks. The first part addresses the challenges of parameter uncertainties (e.g., payload weight) in heavy port handling equipment. We propose a robust strategy that formholds the task as a stochastic optimal control problem. By employing Polynomial Chaos Expansion (PCE) to quantify uncertainties and an adaptive sequential convex optimization (SCvx) algorithm for efficient solving, the proposed method enhances computational efficiency by over 45% compared to advanced pseudospectral methods while ensuring high consistency with Monte Carlo simulations. The second part extends these principles to the autonomous berthing of Unmanned Surface Vehicles (USVs) in complex harbor environments. To overcome the high computational costs of online nonlinear programming, we develop a hierarchical supervised-learning framework. This approach utilizes offline expert trajectories—generated via a corridor-constrained optimization method (AGHA*+SSC)—to train a two-stage neural planner. The resulting system enables millisecond-level state-feedback control and maintains safe maneuvering even under 15% model parameter perturbations. Together, these studies demonstrate the synergy between rigorous mathematical optimization and the rapid inference capabilities of deep learning for resilient maritime operations.