{"id":"arxiv:2606.00252","kind":"preprint","title":"HOIST: Humanoid Optimization with Imitation and Sample-efficient Tuning for Manipulating Suspended Loads","authors":["Songyang Liu","Shunyu Yao","Dingyuan Huang","Shuai Li"],"author_count":4,"year":2026,"published":"2026-05-29","venue":null,"cited_by":0,"doi":null,"arxiv_id":"2606.00252","nct_id":null,"source_url":"https://arxiv.org/abs/2606.00252","archive_url":"https://space-kitty.com/papers/arxiv:2606.00252","updated":"2026-05-29","pmid":null,"pmcid":null,"openalex_id":"W7163198049","pdf_url":"https://arxiv.org/pdf/2606.00252","abstract":"Manipulating suspended payloads with humanoid robots is challenging because the robot can only influence an underactuated, oscillatory load through whole-body motion and intermittent contact. Imitation learning provides safe initial behavior but does not directly optimize final placement, while reinforcement learning from scratch is unsafe and sample-inefficient on real humanoids. We present HOIST-Humanoid Optimized with Imitation and Sample-efficient Tuning for manipulating suspended loads. HOIST first finetunes a high-level vision-language-action (VLA) policy from virtual-reality (VR) teleoperation demonstrations and executes its commands through a whole-body controller. It then uses VLA rollouts and iterative batched RL to improve placement accuracy and stopping behavior. Experiments in simulation and on a real humanoid show that HOIST improves over imitation-only and additional-demonstration baselines; compared with pure VLA rollouts, HOIST reduces translational placement error by 19.9 cm and raw angular error by 3.56 degrees, demonstrating the potential of humanoids for underactuated material-handling tasks.","abstract_license":"CC0-1.0 (arXiv descriptive metadata)","license":null,"is_open_access":false,"retracted":false,"peer_reviewed":false,"categories":["cs.RO","cs.LG"],"details":null,"provenance":"arxiv","attribution":"Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.","topics":[{"slug":"robotics-embodied","name":"Robot Learning & Embodied AI","domain_slug":"ai"}],"cited_in_dossiers":[]}