Scalable Behavior Cloning with Open Data, Training, and Evaluation

Scalable Behavior Cloning with Open Data, Training, and Evaluation

Featured

Arthur Allshire, Himanshu Gaurav Singh, Ritvik Singh, Adam Rashid, Hongsuk Choi, David McAllister, Justin Yu, Yiyuan Chen, Huang Huang, Pieter Abbeel, Xi Chen, Rocky Duan, Phillip Isola, Jitendra Malik, Fred Shentu, Guanya Shi, Philipp Wu, Angjoo Kanazawa · · 2026

Framework

Python

License

Apache-2.0

Stars

200

Summary

We introduce ABC, a fully open-source stack for manipulation with behavior cloning. At its core is ABC-130K: the largest open-source teleoperation dataset to date, featuring 3,500 hours of data spanning over 130K episodes across 195 diverse tasks. Furthermore, we open-source o...

Abstract Summary

We introduce ABC, a fully open-source stack for manipulation with behavior cloning. At its core is ABC-130K: the largest open-source teleoperation dataset to date, featuring 3,500 hours of data spanning over 130K episodes across 195 diverse tasks. Furthermore, we open-source our accessible hardware setup, training infrastructure, and simulation pipeline. We also release 400 hours of sim-teleop data and provide a co-training recipe that produces correlated simulation and real-world evaluation, offering a reliable proxy for ablating model-design and training decisions before costly real-world evaluation. We explore various training recipes and compare common architectural choices for Diffusion Transformers (DiT) and Vision-Language-Action (VLA) models, grounding our findings in real-world evaluations. The resulting policies successfully execute dexterous tasks such as box folding and extracting credit cards from wallets. By providing a reproducible toolkit, we aim to place researchers on an equal footing, establishing the necessary foundation to learn the ABCs of Behavior Cloning together as a community.

Key Points

  • We introduce ABC, a fully open-source stack for manipulation with behavior cloning. At its core is ABC-130K: the largest open-source teleoperation dat...

Abstract

We introduce ABC, a fully open-source stack for manipulation with behavior cloning. At its core is ABC-130K: the largest open-source teleoperation dataset to date, featuring 3,500 hours of data spanning over 130K episodes across 195 diverse tasks. Furthermore, we open-source our accessible hardware setup, training infrastructure, and simulation pipeline. We also release 400 hours of sim-teleop data and provide a co-training recipe that produces correlated simulation and real-world evaluation, offering a reliable proxy for ablating model-design and training decisions before costly real-world evaluation. We explore various training recipes and compare common architectural choices for Diffusion Transformers (DiT) and Vision-Language-Action (VLA) models, grounding our findings in real-world evaluations. The resulting policies successfully execute dexterous tasks such as box folding and extracting credit cards from wallets. By providing a reproducible toolkit, we aim to place researchers on an equal footing, establishing the necessary foundation to learn the ABCs of Behavior Cloning together as a community.

Share

Related Papers

LA4VLA: Learning to Act without Seeing via Language-Action Pretraining

LA4VLA: Learning to Act without Seeing via Language-Action Pretraining

Tao Lin, Yuxin Du, Yiran Mao et al. · arXiv · Jun 2026

Vision-Language-Action (VLA) models are commonly pretrained on robot demonstrations by jointly mapping visual observations and language instructions to actions. However, dense visual-action supervision can dominate the comparatively sparse language-action signal. As a result, ...

Manipulation VLA Models Reinforcement Learning Imitation Learning Sensing & Perception
PDF Intermediate
No code repo Code updated: Jun 2026
E-TTS: A New Embodied Test-Time Scaling Framework for Robotic Manipulation

E-TTS: A New Embodied Test-Time Scaling Framework for Robotic Manipulation

Wen Ye, Peiyan Li, Tingyu Yuan et al. · arXiv · Jun 2026

Recently, a few works have made early attempts to study test-time scaling for embodied tasks. However, two major challenges remain unsolved: (1) reasoning can effectively improve the performance of the policy, but its scaling mechanism has seldom been studied; (2) historical i...

Manipulation VLA Models Reinforcement Learning Sensing & Perception
PDF Intermediate
No code repo Code updated: Jun 2026
Hand-in-the-Loop: Improving Dexterous VLA via Seamless Interventional Correction

Hand-in-the-Loop: Improving Dexterous VLA via Seamless Interventional Correction

Zhuohang Li, Liqun Huang, Wei Xu et al. · arXiv · May 2026

Vision-Language-Action (VLA) models are prone to compounding errors in dexterous manipulation, where high-dimensional action spaces and contact-rich dynamics amplify small policy deviations over long horizons. While Interactive Imitation Learning (IIL) can refine policies through human takeover data...

Manipulation Imitation Learning VLA Models Sensing & Perception
PDF Advanced
No code repo Code updated: May 2026
HumanoidUMI: Bridging Robot-Free Demonstrations and Humanoid Whole-Body Manipulation

HumanoidUMI: Bridging Robot-Free Demonstrations and Humanoid Whole-Body Manipulation

Hongwu Wang, Chenhao Yu, Youhao Hu et al. · arXiv · Jun 2026

High-quality demonstration data are essential for humanoid robot skill learning, especially for whole-body behaviors that require coordinated perception, locomotion, and manipulation. Existing data-collection methods largely rely on robot teleoperation, which is constrained by...

Humanoids Manipulation Reinforcement Learning Imitation Learning Locomotion & Navigation Sensing & Perception
Code PDF Intermediate
GitHub ★ 0 Code updated: Jun 2026