Event
LiDAR-Based Label-Free Trajectory Path Correction Framework Proposed
1 reports · 1 sources
Yilong Chen and Yong K. Cho propose a LiDAR-only real-time framework that learns interaction-aware path correction from unlabeled demonstrations, targeting construction robots. The method uses constant-velocity extrapolation to generate residual learning targets, with a GNN modeling the relationships among the robot, pedestrians, and goal context; at deployment, an extended Kalman filter tracks LiDAR detections to build a dynamic graph, and the framework offsets planner-generated waypoints through a planner-agnostic interface. In bricklaying trials, the GNN reduced path irregularity by 39.2% and improved the minimum TTC. Ablation experiments support a 1.0-second prediction horizon and confirm the roles of goal context, reflection augmentation, and feature normalization.
Report timeline
- LiDAR-based GNN framework learns interaction-aware path corrections from unlabeled trajectories for construction robots
The study proposes a real-time, LiDAR-only framework that learns interaction-aware path corrections from unlabeled demonstrations: constant-velocity extrapolation generates residual learning targets, a GNN models relationships among the robot, pedestrians, and goal context, and at deployment an extended Kalman filter tracks LiDAR detections to build dynamic graphs, with the GNN offsetting planner-generated waypoints through a planner-agnostic interface.
arXiv cs.RO ·