RoboPARA: Dual-Arm Robot Planning with Parallel Allocation and Recomposition Across Tasks
Shiying Duan, Pei Ren, Nanxiang Jiang, Zhengping Che, Jian Tang, Zhaoxin Fan, Yifan Sun, Wenjun Wu
TL;DR
RoboPARA introduces a novel LLM-driven framework for dual-arm robot task planning that enhances parallelism and efficiency through dependency graph modeling and optimized traversal, supported by a new evaluation dataset.
Contribution
The paper presents RoboPARA, a two-stage planning framework utilizing dependency graphs and graph re-traversal, along with the X-DAPT dataset for evaluating dual-arm task parallelism.
Findings
RoboPARA outperforms existing methods in efficiency and reliability.
The framework effectively maximizes task parallelism in complex scenarios.
The X-DAPT dataset enables comprehensive evaluation across diverse tasks.
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…
Peer Reviews
Decision·ICLR 2026 Poster
1.The two-stage architecture of RoboPARA effectively models task dependencies and optimizes dual-arm parallelism, fully exploiting the collaborative potential of dual-arm robots. 2.The X-DAPT dataset is the first dedicated to evaluating dual-arm task parallelism, covering diverse scenarios and difficulty levels, providing a comprehensive benchmark. 3.RoboPARA demonstrates excellent performance in experiments, with significant improvements in parallel steps, execution time reduction, and task s
1.Limited generalization capability: RoboPARA relies on predefined skill libraries and scenario templates, requiring abstraction for novel tasks or environments. This hinders its scalability to long-horizon, multi-stage and complex tasks and out-of-distribution scenes. 2.Mismatch between optimization objective and real-robot deployment: The planning stage minimizes estimated action duration, but accurate execution latency is hard to obtain and estimate accurately in real-world settings. As a re
The paper's primary strength is its formalization and direct confrontation of parallelism in dual-arm manipulation. While prior work often results in sequential execution , this paper defines a new, relevant problem ("Dual-Arm Cooperative Scheduling") and proposes a solution explicitly designed to optimize for it. This focus on decoupling tasks for parallel execution, rather than purely sequential or fully synchronous collaboration, is a significant and practical contribution. The two-stage, hy
The authors designed both the solution (RoboPARA) and the primary benchmark (X-DAPT) on which it demonstrates overwhelming superiority. The results in Tables 1 and 2 show that RoboPARA achieves high scores on the new parallelism metrics (PPR and APR), while all seven baseline methods score 0.000 or near-zero on these metrics in almost every single category. This result is suggesting that the X-DAPT benchmark is overtuned to the specific graph-based, parallel-aware architecture of RoboPARA. This
1. While minimizing task time or improving efficiency in dual-arm robots is not a new problem, the paper’s approach is novel in leveraging the semantic reasoning capability of LLMs to extract task dependency graphs from complex multitasking instructions to maximize parallelism. 2. Once released, the proposed dataset could make a meaningful contribution as a dual-arm task dependency dataset, potentially serving as a valuable benchmark for studying parallel manipulation and LLM-based task plannin
1. While the paper includes a DAG validation and correction step to ensure logical consistency, it seems that this process lacks deep physics-aware validation. Crucial physical constraints like spatial conflicts, reachability, and precise timing are not rigorously checked during the initial DAG generation or correction phase, potentially leading to logically sound but physically infeasible plans that might only be detected later during execution or scheduling. 2. The paper introduces a rollback
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsReinforcement Learning in Robotics · Social Robot Interaction and HRI · Multimodal Machine Learning Applications