A careful examination of large behavior models for multitask dexterous manipulation
Science Robotics · 15.04.2026 · 11 (113)
DOI: 10.1126/scirobotics.aea6201 ↗Kurz gefasst
Fachartikel in der Zeitschrift Science Robotics, erschienen am 15.04.2026. Verfasst von Jose Barreiros und 59 weiteren Autor:innen. Beteiligte Einrichtung: Toyota Research Institute, Cambridge, MA 02139, USA. Die Arbeit wurde bisher 13-mal zitiert (Crossref) und nennt 85 Literaturangaben. Gefördert von: Toyota Research Institute.
Abstract (Originalsprache)
Robot manipulation has seen tremendous progress in recent years, with imitation learning policies enabling successful performance of dexterous and hard-to-model tasks. Concurrently, scaling data and model size has led to the development of capable language and vision foundation models, motivating large-scale efforts to create general-purpose robot foundation models. Although these models have garnered considerable enthusiasm and investment, meaningful evaluation of real-world performance remains a challenge, limiting the pace of development and inhibiting a nuanced understanding of current capabilities. Here, we rigorously evaluated multitask robot manipulation policies, referred to as large behavior models, by extending the diffusion policy paradigm across a corpus of simulated and real-world robot data. We proposed and validated an evaluation pipeline to rigorously analyze the capabilities of these models with statistical confidence. We compared against single-task baselines through blind, randomized trials in a controlled setting, using both simulation and real-world experiments. We found that multitask pretraining made the policies more successful and robust and enabled teaching complex new tasks more quickly, using a fraction of the data when compared with single-task baselines. Moreover, performance predictably increased as pretraining scale and diversity grows.
Autor:innen und Institutionen
- Toyota Research Institute, Cambridge, MA 02139, USA.
- Cornell University, Ithaca, NY 14850, USA.
- Massachusetts Institute of Technology, Cambridge, MA 02139, USA.
ORCID-Profile: Jose Barreiros, Andrew Beaulieu, Aditya Bhat, Rick Cory, Eric Cousineau, Hongkai Dai, Ching-Hsin Fang, Kunimatsu Hashimoto, Muhammad Zubair Irshad, Masha Itkina, Naveen Kuppuswamy, Kuan-Hui Lee, Katherine Liu, Dale McConachie, Ian McMahon, Haruki Nishimura, Calder Phillips-Grafflin, Charles Richter, Paarth Shah, Krishnan Srinivasan, Blake Wulfe, Chen Xu, Mengchao Zhang, Alex Alspach, Maya Angeles, Alejandro Castro, Richard Denitto, Emma Dixon, Allison Henry, Joseph Masterjohn, Masayuki Masuda, Patrick Miller, Paul Mitiguy, Yuki Noguchi, Aykut Onol
Förderung
- Toyota Research Institute
Zitieren
APA
Barreiros, J., Beaulieu, A., Bhat, A., Cory, R., Cousineau, E., Dai, H., Fang, C. H., Hashimoto, K., Irshad, M. Z., Itkina, M., Kuppuswamy, N., Lee, K. H., Liu, K., McConachie, D., McMahon, I., Nishimura, H., Phillips-Grafflin, C., Richter, C., Shah, P., …, & Onol, A. (2026). A careful examination of large behavior models for multitask dexterous manipulation. Science Robotics, 11(113). https://doi.org/10.1126/scirobotics.aea6201
Vancouver
Barreiros J, Beaulieu A, Bhat A, Cory R, Cousineau E, Dai H, et al. A careful examination of large behavior models for multitask dexterous manipulation. Science Robotics. 2026;11(113). doi:10.1126/scirobotics.aea6201
BibTeX
@article{Barreiros2026,
author = {Barreiros, Jose and Beaulieu, Andrew and Bhat, Aditya and Cory, Rick and Cousineau, Eric and Dai, Hongkai and Fang, Ching-Hsin and Hashimoto, Kunimatsu and Irshad, Muhammad Zubair and Itkina, Masha and Kuppuswamy, Naveen and Lee, Kuan-Hui and Liu, Katherine and McConachie, Dale and McMahon, Ian and Nishimura, Haruki and Phillips-Grafflin, Calder and Richter, Charles and Shah, Paarth and Srinivasan, Krishnan and Wulfe, Blake and Xu, Chen and Zhang, Mengchao and Alspach, Alex and Angeles, Maya and Arora, Kushal and Guizilini, Vitor Campagnolo and Castro, Alejandro and Chen, Dian and Chu, Ting-Sheng and Creasey, Sam and Curtis, Sean and Denitto, Richard and Dixon, Emma and Dusel, Eric and Ferreira, Matthew and Goncalves, Aimee and Gould, Grant and Guoy, Damrong and Gupta, Swati and Han, Xuchen and Hatch, Kyle and Hathaway, Brendan and Henry, Allison and Hochsztein, Hillel and Horgan, Phoebe and Iwase, Shun and Jackson, Donovon and Karamcheti, Siddharth and Keh, Sedrick and Masterjohn, Joseph and Masuda, Masayuki and Mercat, Jean and Miller, Patrick and Mitiguy, Paul and Nguyen, Tony and Nimmer, Jeremy and Noguchi, Yuki and Ong, Reko and Onol, Aykut},
title = {A careful examination of large behavior models for multitask dexterous manipulation},
journal = {Science Robotics},
year = {2026},
volume = {11},
number = {113},
doi = {10.1126/scirobotics.aea6201}
}
EUVIMED