RoboPARA: Dual-Arm Robot Planning with Parallel Allocation and Recomposition Across Tasks

TL;DR

RoboPARA, a novel large language model (LLM)-driven framework for dual-arm task parallelism planning, is proposed and the Cross-Scenario Dual-Arm Parallel Task dataset (X-DAPT dataset) is introduced, the first dataset specifically designed to evaluate dual-arm task parallelism across diverse scenarios and difficulty levels.

Abstract

Dual-arm robots play a crucial role in improving efficiency and flexibility in complex multitasking scenarios.While existing methods have achieved promising results in task planning, they often fail to fully optimize task parallelism, limiting the potential of dual-arm collaboration.To address this issue, we propose RoboPARA, a novel large language model (LLM)-driven framework for dual-arm task parallelism planning.RoboPARA employs a two-stage process: (1) Dependency Graph-based Planning Candidates Generation, which constructs directed acyclic graphs (DAGs) to model task dependencies and eliminate redundancy, and (2) Graph Re-Traversal-based Dual-Arm Parallel Planning, which optimizes DAG traversal to maximize parallelism while maintaining task coherence.In addition, we introduce the Cross-Scenario Dual-Arm Parallel Task dataset (X-DAPT dataset), the first dataset specifically designed to evaluate dual-arm task parallelism across diverse scenarios and difficulty levels.Extensive experiments demonstrate that RoboPARA significantly outperforms existing planning methods, achieving higher efficiency and reliability, particularly in complex task combinations.Our code is publicly available at https://github.com/AiDuanshiying/RoboPARA.

Figures (15)

• Figure 1: RoboPARA enables efficient parallel manipulation and collaborative dual-arm execution, resulting in a 30% to 50% reduction in execution time while maintaining task success. When deployed on a humanoid robot, it demonstrates behaviors closely aligned with human activities.
• Figure 2: The RoboPARA framework. Given user instructions, our two-stage framework outputs a dual-arm execution schedule. Stage 1 generates dependency graph-based planning candidates: it builds an LLM‑based DAG from experiences and iteratively refines it via error correction (Appendix Alg. \ref{['very_DAG']}) and updates. Stage 2 re-traverses the graph to plan parallel execution: it identifies parallelizable tasks and resolves deadlocks (Appendix Alg. \ref{['scheduler_alg']} and Alg. \ref{['alg:choose_arm']}), then finalizes task assignment.
• Figure 3: Statistical evaluation of X-DAPT dataset. The pie charts show the appearing skills (outer circle) and required number of arms (inner circle) of the 10 scenes. The bar chart shows the percentage of instructions with different action steps. X-DAPT is larger in scale, richer in scenes, and more diverse in skills than previous datasets Embodied_Task_Planning_with_Large_Language_ModelsALFRED. It is closely aligned with daily life and specifically designed to highlight parallel execution.
• Figure 4: Real-world deployment comparison of RoboPARA and baselines on Franka Research 3 robot in an agricultural greenhouse scene. Voyager is excluded because it cannot provide explicit plans for each arm. See Appendix Sec. \ref{['franka_real_test_app']} for details and specific plans regarding this field test.
• Figure 5: The percentage of packages with different numbers of implementation actions.