Enable dexterous manipulation within generalist robotic manipulation policies

Develop learning methods and control strategies that allow generalist robotic manipulation policies to reliably handle dexterous, contact-rich, and coordinated bimanual manipulation tasks.

Background

The evaluation suite in the paper includes long-horizon, dexterous tasks (e.g., tool use, precise manipulation). While the authors demonstrate improvements from multitask pretraining, they identify handling dexterity as an open problem for generalist policies.

Advances here would broaden the range of real-world tasks that LBMs can perform autonomously and robustly.

References

Despite progress in training generalist policies, challenges such as catastrophic forgetting, data heterogeneity, scarcity of high-quality data, multimodal fusion, handling dexterity, and maintaining real-time inference speed remain open research problems.

— A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation  (2507.05331 - Team et al., 7 Jul 2025) in Section 2.1, Related Work—Robot Learning at Scale

Whether such trajectories improve downstream policy learning remains for future work.

— DreamHand: Repurposing Video Diffusion Models for Occlusion-Robust Egocentric 3D Hand Motion Recovery  (2608.20308 - Liu et al., 20 Aug 2026) in Appendix, Section Qualitative Results and Application, subsection “Retargeting to a Dexterous Hand” (Section sec_supp_qualapp)