Autonomous driving technology has the potential to improve transportation efficiency and safety, extending beyond traditional road vehicles to include low-speed mobility aids in unstructured environments like airport terminals and shopping malls. In these crowded shared spaces, the interaction between pedestrians and autonomous vehicles plays a critical role in designing effective navigation algorithms. This research focuses on developing a coupled prediction and planning algorithm for low-speed autonomous vehicles in unstructured environments. Leveraging trajectory data from shared spaces, we employ a data-driven approach to train a pedestrian trajectory predictor. Notably, our approach highlights an innovative method for modeling in-between interaction effects on pedestrian motion. The pedestrian trajectory predictor's output is incorporated into both a Deep Reinforcement Learning (DRL)-based and a Model Predictive Control (MPC)-based motion planner. Comparative analysis provides insights into the potential effectiveness of integrating these methods for a safer and more efficient navigation algorithm. Addressing social aspects, the motion planner considers pedestrian comfort based on personal space constraints and other insights gathered from
an online user study. This interdisciplinary algorithm advances autonomous navigation by merging motion planning and human-robot interaction, emphasizing efficiency and pedestrian safety in shared spaces


Mahsa Golchoubian, PhD candidate in Systems Design Engineering

Attend in person in E7-6443 or on Zoom

Zoom link:
Meeting ID: 941 8603 1942
Passcode: 230362

Attending this seminar will count towards the graduate student seminar attendance milestone!