Benchmarking Generalizable Bimanual Manipulation: RoboTwin Dual-Arm Collaboration Challenge at CVPR 2025 MEIS Workshop

2025-06-29

AI Summary

This work addresses the challenge of generalizable dual-arm robotic manipulation, targeting robust handling of rigid, deformable, and tactile-sensitive objects. Method: We propose a three-stage competition paradigm and introduce the first multi-physical dual-arm benchmark suite—comprising 17 complex tasks—designed to evaluate generalization across object properties. The benchmark is validated concurrently in RoboTwin (1.0/2.0) simulation and on the AgileX COBOT-Magic physical platform. We systematically assess dual-arm generalization strategies, integrating state-of-the-art methods including SEM and AnchorDP3. Contribution/Results: The competition attracted 64 international teams (>400 participants), yielding multiple high-performance solutions. It establishes the first open-source, cross-modal, simulation-to-real dual-arm generalization benchmark, providing both methodological frameworks and empirical foundations for embodied dual-arm collaborative learning.

Technology Category

• Intelligent Robots: Manipulation
• Natural Language Processing: Safety and Robustness
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Embodied Artificial Intelligence (Embodied AI) is an emerging frontier in robotics, driven by the need for autonomous systems that can perceive, reason, and act in complex physical environments. While single-arm systems have shown strong task performance, collaborative dual-arm systems are essential for handling more intricate tasks involving rigid, deformable, and tactile-sensitive objects. To advance this goal, we launched the RoboTwin Dual-Arm Collaboration Challenge at the 2nd MEIS Workshop, CVPR 2025. Built on the RoboTwin Simulation platform (1.0 and 2.0) and the AgileX COBOT-Magic Robot platform, the competition consisted of three stages: Simulation Round 1, Simulation Round 2, and a final Real-World Round. Participants totally tackled 17 dual-arm manipulation tasks, covering rigid, deformable, and tactile-based scenarios. The challenge attracted 64 global teams and over 400 participants, producing top-performing solutions like SEM and AnchorDP3 and generating valuable insights into generalizable bimanual policy learning. This report outlines the competition setup, task design, evaluation methodology, key findings and future direction, aiming to support future research on robust and generalizable bimanual manipulation policies. The Challenge Webpage is available at https://robotwin-benchmark.github.io/cvpr-2025-challenge/.

Problem

• Advancing generalizable bimanual manipulation for complex tasks
• Benchmarking dual-arm systems in rigid and deformable object handling
• Developing robust policies for tactile-sensitive object collaboration

Innovation

• RoboTwin Simulation platform for dual-arm tasks
• AgileX COBOT-Magic Robot for real-world testing
• 17 diverse bimanual manipulation scenarios evaluated