Skip to main content Skip to footer
RESEARCH PROJECTS & TEAM

Robot Decision & Control

Combines learning from demonstration and Low-Code task planning to lower programming barriers, enabling efficient multi-robot collaborative scheduling.

Robot Decision & Control

Related
Technology
Learning from Demonstration AMR Precision Navigation & Localization Low-Code Task Planning Multi-Robot Collaborative Scheduling
Real-time Scheduling and Path Planning for AMR

Research Project

Real-time Scheduling and Path Planning for AMR

The subproject addresses the challenges of real-time scheduling and path planning for AMR fleets in dynamic environments by proposing an intelligent decision-making framework that integrates GNN, DDQN, genetic algorithms, and Coarse-to-Fine path optimization. To cope with the uncertainty caused by dynamic job arrivals, the proposed system will employ GNN to extract real-time representations and states of the operating environment. We further use DDQN to learn the optimal timing for triggering the job scheduler, replacing conventional fixed-period scheduling. This approach reduces unnecessary computational overhead and avoids delays in scheduling responses, thereby enhancing the system's dynamic resilience. To improve the matching efficiency between AMRs and tasks, this study introduces a genetic algorithm to accelerate the solution of large-scale combinatorial optimization problems. By shortening the convergence time of the task-matching algorithm, the system can reduce AMR idle and waiting ratios, further improving the operational throughput of the entire fleet. Finally, for path planning, we adopt a two-stage Coarse-to-Fine path optimization architecture. The first stage rapidly generates reference trajectories, while the second stage incorporates spatiotemporal constraints to refine and correct the paths. This method can effectively reduce the computational burden of complex path simulations and improve the system's real-time responsiveness and planning efficiency under environmental changes. Overall, this study aims to strengthen the scheduling efficiency, computational performance, and system resilience of AMR fleets in dynamic task environments through real-time state perception, intelligent scheduling triggers, efficient task matching, and hierarchical path optimization.

Project Details

Principal Investigator

Yu-Lun Huang Associate Professor

PI Details