Multi-Robot Vision-Based Task and Motion Planning for EV Battery Disassembly and Sorting

2025-09-25

AI Summary

To address low coordination accuracy, motion redundancy, and high collision risk in multi-robot disassembly of electric vehicle batteries under cluttered, dynamic environments, this paper proposes a four-layer architecture integrating visual perception and task-motion co-planning. It combines symbolic task planning with TP-GMM-based imitation learning to jointly optimize task assignment—considering cost and reachability—and generate compact, collision-free trajectories. The system integrates YOLOv8-based stereo localization, OctoMap for real-time 3D mapping, FCL for collision checking, MoveIt for motion planning, and digital twin–enabled predictive control to support online visual obstacle avoidance. Experiments on a dual-UR10e platform demonstrate that, compared to the RRTConnect baseline, the proposed method reduces end-effector path length by 63.3%, decreases total task completion time by 8.1%, and significantly lowers inter-arm motion volume and spatial overlap—by 47% for the latter—validating its superior efficiency, safety, and coordination performance.

Technology Category

• Intelligent Robots: Motion and Path Planning
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Electric-vehicle (EV) battery disassembly requires precise multi-robot coordination, short and reliable motions, and robust collision safety in cluttered, dynamic scenes. We propose a four-layer task-and-motion planning (TAMP) framework that couples symbolic task planning and cost- and accessibility-aware allocation with a TP-GMM-guided motion planner learned from demonstrations. Stereo vision with YOLOv8 provides real-time component localization, while OctoMap-based 3D mapping and FCL(Flexible Collision Library) checks in MoveIt unify predictive digital-twin collision checking with reactive, vision-based avoidance. Validated on two UR10e robots across cable, busbar, service plug, and three leaf-cell removals, the approach yields substantially more compact and safer motions than a default RRTConnect baseline under identical perception and task assignments: average end-effector path length drops by $-63.3%$ and makespan by $-8.1%$; per-arm swept volumes shrink (R1: $0.583 ightarrow0.139,mathrm{m}^3$; R2: $0.696 ightarrow0.252,mathrm{m}^3$), and mutual overlap decreases by $47%$ ($0.064 ightarrow0.034,mathrm{m}^3$). These results highlight improved autonomy, precision, and safety for multi-robot EV battery disassembly in unstructured, dynamic environments.

Problem

• Developing multi-robot coordination for EV battery disassembly in cluttered environments
• Ensuring collision safety and reliable motions in dynamic disassembly scenes
• Integrating task planning with vision-based perception for precise component handling

Innovation

• Four-layer TAMP framework with task planning and motion allocation
• Stereo vision and YOLOv8 for real-time component localization
• OctoMap-based 3D mapping with FCL collision checking in MoveIt