MAD Games workshop on Multi-Agent Dynamic Games at ICRA 2024 was organized by Rahul Mangharam, Hongrui Zheng, Shuo Yang, Johannes Betz and Venkat Krovi. icra2024-madga...
This talk focuses on two key aspects of the multi-agent motion planning problem in mixed environments. First, the majority of the traditional multi-agent planning algorithms rely on simplifying assumptions about the robot dynamics, resulting in unreliable execution under realistic conditions. We introduce a scalable and efficient algorithm called Conflict-Based MPC (CB-MPC), that plans dynamically feasible trajectories for multiple robots, while efficiently resolving conflicts using a CBS-like conflict-tree. Second, in many applications like autonomous driving and sidewalk delivery, the intentions of other agents in the environment are uncontrollable and uncertain. A successful controller in these environments has to simultaneously reason about discrete intent uncertainty as well as continuous observation uncertainty. We present an MPC-based planner that leverages a novel policy parameterization to efficiently reason about multi-modal uncertainties. We show that this parametrization increases feasibility and improves navigation performance in mixed environments.
Ardalan Tajbakhsh is a PhD candidate at Carnegie Mellon University focusing on dynamic multi-agent planning and control for multi-agent robotics applications. Prior to his PhD, he was working as a Planning and Control Engineer at Zebra Technologies in the advanced development team. His main focus was leading the algorithm development, testing, and deployment efforts for planning and execution of multi-agent systems in warehouses.
Негізгі бет ICRA 2024 - MAD Games Workshop - Ardalan Tajbakhsh
Пікірлер