M.Sc Thesis
M.Sc Thesis
Human-Aware Robot Navigation in Dynamic Crowds Using Attention-Based Interaction Models
U.K Roy
Abstract
This study addresses the challenge of robot navigation in crowded environments by incorporating human group modeling into the navigation system. Existing models primarily focus on individual human-agent interactions, neglecting the importance of group dynamics in dynamic crowds. To bridge this gap, we develop a crowd navigation framework that integrates human group modeling and evaluates several baselines. As part of this framework, we propose a reactive-based model, GAvoid, which uses tangent-based actions to mitigate the issues associated with group avoidance, incorporating it alongside other baseline models. While GAvoid demonstrates improvements in group collision rates and does not significantly reduce the success rate, its reactive nature leads to an increase in individual collision rates, as it focuses solely on group behaviors. Additionally, in several scenarios involving multiple close groups, GAvoid fails to effectively avoid groups, revealing significant limitations in its effectiveness. To overcome these challenges, we develop an attention-based system, GRAM (Group-Responsive Attention Mechanism), which achieves the highest success rate of 82\%, significantly outperforming the best-performing baseline method, which only achieves 57\%. This demonstrates GRAM's ability to effectively balance success rate, group collision rate, and individual human collision rate, establishing it as a robust solution for group-aware robot navigation in dynamic crowds.
News
March 1, 2025: Exciting News: Our First Paper Submitted to IROS 2025
We are thrilled to announce that our first research paper, derived from Chapter 3 of my thesis on Human-Aware Robot Navigation Using Group Modeling, has been submitted to the prestigious IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025.
Here you can find the details of our submitted work: click here
March 2, 2025: Successfully defended my M.Sc. thesis, "Human-Aware Robot Navigation in Dynamic Crowds Using Reinforcement Learning and Attention-Based Interaction Models," which focuses on developing socially compliant navigation strategies for robots in dynamic human environments.
Please be patient for more details—full information will be shared soon as the work progresses toward publication!
Please note that not all details of this project are shared as the work is yet to be published. However, you are welcome to explore and enjoy the simulation results provided here.