Rui Pan

Topic

Model Predictive Control for Robotic Manipulators in Dynamic Environments: Dynamic Obstacle Avoidance and Moving-Target Visual Servoing

Department of Mechanical Engineering

Date & location

• Thursday, October 8, 2026

• 12:00 P.M.

• Engineering Office Wing, Room 430

Reviewers

Supervisory Committee

• Dr. Yang Shi, Department of Mechanical Engineering, University of Victoria (Supervisor)

• Dr. Daniela Constantienscu, Department of Mechanical Engineering, UVic (Member) 

External Examiner

• Dr. Xiaodai Dong, Department of Electrical and Computer Engineering, University of Victoria 

Chair of Oral Examination

• Dr. Simon Devereaux, Department of History, UVic

   

Abstract

Robotic manipulators operating in dynamic environments require control strategies that can generate control actions online while satisfying safety and operational constraints. This thesis investigates model predictive control (MPC) frameworks for robotic manipulators in dynamic environments, with a focus on dynamic obstacle avoidance and moving-target visual servoing. In these tasks, the motion of the obstacle or target is not assumed to be perfectly known. Online estimation is integrated with MPC so that newly acquired sensing information can be incorporated into the MPC optimization problem while physical constraints are explicitly enforced. 

First, this thesis develops an MPC-based dynamic obstacle avoidance framework for a UR10e manipulator. The manipulator geometry is represented by a set of collision-checking points, allowing obstacle avoidance, self-collision, and ground clearance constraints to be formulated. The position measurements of the obstacle provided by the RGB-D camera are noisy. A Kalman filter is then used to estimate the obstacle position and velocity, and the estimated obstacle state is used to predict obstacle motion over the MPC prediction horizon. Simulation results show that the Kalman filter improves the accuracy of obstacle-position estimation and provides a reliable estimate of obstacle velocity. The proposed control framework drives the end effector toward the target while maintaining safe clearance from the moving obstacle and satisfying the prescribed safety and joint constraints.

Second, this thesis develops a disturbance-rejection-based robust MPC (RMPC) framework for visual servoing of a moving target with unknown velocity. To this end, a virtual camera and virtual image-moment features are introduced to derive visual feature dynamics, in which the target velocity appears as an unknown exogenous input. A nonlinear disturbance observer is then designed to estimate this exogenous input online, and the residual estimation error bound is incorporated into the RMPC constraint-tightening design. Sufficient conditions are established for robust constraint satisfaction, recursive feasibility, and regional practical stability. Experiments on a UR10e manipulator equipped with an eye-in-hand RGB-D camera show that the proposed framework improves moving-target tracking performance while maintaining bounded tracking errors and joint velocities. 

Together, these studies demonstrate that MPC provides an effective framework for online robotic manipulation in dynamic environments. In the proposed frameworks, online estimates of the obstacle state or target velocity are incorporated into the MPC design. This allows the controller to adapt its control actions to newly acquired sensing information while explicitly enforcing safety and operational constraints.