DuoBench: A Reproducible Benchmark for Bimanual Manipulation in Simulation and the Real World

2026-06-10 · Source: Artificial Intelligence · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

DuoBench is an extensible benchmarking framework designed for bimanual manipulation policies, specifically on the FR3 Duo robot platform. This framework features eleven distinct tasks categorized into four coordination types, implemented in both simulation and partially in real-world environments using reproducible task recipes and 3D-printable assets. It introduces a stage-based evaluation scheme to enable fine-grained semantic failure analysis, moving beyond simple binary success metrics. Additionally, DuoBench provides human-teleoperated datasets for all benchmark tasks. Initial evaluations of several dual-arm imitation-learning and vision-language-action policies demonstrated that current methods struggle with bimanual manipulation, particularly during early interaction stages, parallel arm execution, and effective transfer between simulated and real-world settings. DuoBench aims to serve as a reproducible testbed for diagnosing these challenges and advancing dual-arm policy learning.

Key takeaway

For Robotics Engineers developing bimanual manipulation policies, DuoBench provides a critical tool for evaluating and diagnosing system performance. You should utilize its stage-based evaluation and reproducible tasks to pinpoint specific failure modes, especially concerning early interaction and parallel arm coordination. This framework helps you identify where current imitation-learning and vision-language-action policies struggle, guiding your research towards more robust sim-to-real transfer and improved dual-arm control strategies.

Key insights

DuoBench offers a reproducible benchmark for bimanual robot manipulation, revealing current policy limitations in coordination and sim-to-real transfer.

Principles

• Bimanual systems introduce complex control and failure modes.
• Stage-based evaluation enables fine-grained failure analysis.
• Reproducible task recipes aid real-world transfer.

Method

DuoBench implements eleven bimanual tasks across four coordination categories, using reproducible 3D-printable assets and a stage-based evaluation for semantic failure analysis.

In practice

• Test dual-arm policies on FR3 Duo platform.
• Diagnose failures in parallel arm execution.
• Study sim-to-real transfer challenges.

Topics

• Bimanual Manipulation
• Robot Benchmarking
• FR3 Duo Platform
• Simulation-to-Real Transfer
• Imitation Learning
• Robot Control

Best for: Research Scientist, Robotics Engineer, AI Scientist, Machine Learning Engineer

Related on AIssential

• A benchmarking framework for embodied neuromorphic agents — A new framework standardizes benchmarking for neuromorphic control of soft robots in dynamic, real-world environments.
• GroundedPlanBench: Spatially grounded long-horizon task planning for robot manipulation — Jointly planning actions and spatial grounding improves robot task success and action accuracy for VLMs.
• Looking for real world comparisons between WALL OSS pi0.6 and OpenVLA[D] — Deployment friction, not just benchmark scores, dictates the practical utility of robotic manipulation stacks.
• DeVI: Physics-based Dexterous Human-Object Interaction via Synthetic Video Imitation — DeVI enables physics-based dexterous robot control using synthetic 2D videos and a hybrid 2D/3D tracking reward.
• RoboDream: Compositional World Models for Scalable Robot Data Synthesis — A generalizable embodiment-centric world model synthesizes photorealistic robot demonstrations by decoupling motion from environment synthesis.
• RoboLab: A High-Fidelity Simulation Benchmark for Analysis of Task Generalist Policies — RoboLab offers a high-fidelity simulation benchmark to assess robotic policy generalization and sensitivity to external factors.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.