A careful examination of large behavior models for multitask dexterous manipulation

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.

Authors

Institutions

Publication Details

Journal
Science Robotics
Published
2026-04-15
DOI
https://doi.org/10.1126/scirobotics.aea6201
Primary Topic
Robot Manipulation and Learning
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

A careful examination of large behavior models for multitask dexterous manipulation

Damrong Guoy, Muhammad Zubair Irshad, Naveen Kuppuswamy, Hadas Kress‐Gazit et al.
Science Robotics
Robot Manipulation and Learning
article

A careful examination of large behavior models for multitask dexterous manipulation

Damrong Guoy, Muhammad Zubair Irshad, Naveen Kuppuswamy, Hadas Kress‐Gazit, Jean Mercat, Masha Itkina, Andrew Beaulieu, Calder Phillips-Grafflin, Blake Wulfe, Joseph Masterjohn, Sam Creasey, Rareş Ambruş, Mengchao Zhang, Hongkai Dai, Pavel Tokmakov, K Ray Liu, Igor Vasiljevic, Jose Barreiros, Aimee Goncalves, Shun Iwase, Rick Cory, Haruki Nishimura, Ting-Sheng Chu, Siddharth Karamcheti, Aykut ÖNOL, Hillel Hochsztein, Mark Zolotas, Kyle Hatch, Benjamin Burchfiel, Donovon Jackson, Kunimatsu Hashimoto, Michael Sherman, Derick Seale, Alex Alspach, Vitor Campagnolo Guizilini, Eric Cousineau, Jeremy Nimmer, Siyuan Feng, Gordon Richardson, Kushal Arora, Ian McMahon, Reko Ong, Phoebe Horgan, Krishnan Srinivasan, Matthew Tran, Mariah Smith-Jones, Charles Richter, Sean Curtis, Matthew Ferreira, Emma Dixon, Swati Gupta, Basile Van Hoorick, Richard Poyner, Ching-Hsin Fang, Grant Gould, David Tago, Tony Nguyen, Chen Xu, Kerri Fetzer-Borelli, Christopher Rodriguez, Maya Angeles, Allison Henry, Owen Pfannenstiehl, Richard DeNitto, Dale McConachie, Brendan Hathaway, Xuchen Han, Masayuki Masuda, Eric Dusel, Aditya Bhat, Kuan-Hui Lee, Leticia Priebe Mendes Rocha, Paul Mitiguy, Paarth Shah, Russ Tedrake, Sedrick Keh, Dian Chen, Patrick Miller, Sergey Zakharov
article en

Abstract

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.

Science RoboticsVol. 11(113)
Cornell University (US), Toyota Research Institute, Massachusetts Institute of Technology (US)
Toyota Research Institute
Openalex Percentile: Top 20%
Robot Manipulation and Learning
DOISourceOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.