Humanoids learn to cooperatively pick up and carry objects together
Learning Multi-Humanoid Pickup and Transport via Decentralized Object-Centric Control
RoboticsArtificial Intelligence
Summary
Picking up and moving objects together is hard for robots because they have to work closely without getting in each other's way. The authors found a way for multiple humanoid robots to each hold a part of an object and learn how to carry it together without needing special tools or redesigned steps for every task. They showed that skills learned by one robot can help when multiple robots work together. Training them directly in groups improved teamwork. They tested this in computer simulations and on real robots successfully.
What this means in practice
- •For industrial automation teams: Enable multiple humanoid robots to reliably pick up and transport various large or heavy objects cooperatively without redesigning control schemes.
- •For robotics hardware developers: Test and improve real humanoid robot hardware using learned controllers that coordinate multi-robot manipulation in physical environments.
Authors
Bikram Pandit, Mohitvishnu S. Gadde, Aayam Kumar Shrestha, Alan Fern
Abstract
We study cooperative multi-humanoid pickup and transport of objects with varying size, weight, and geometry, requiring robot teams of different sizes. Our approach uses decentralized object-centric control, where each humanoid is assigned a local attachment region on the shared object and learns to realize pickup and transport through gripperless bimanual pinching. This attachment-based interface provides a common control abstraction spanning single-robot pickup, cooperative multi-robot transport, and robot-to-robot handover, without per-task redesign. We find that policies trained only on single-robot pickup already transfer nontrivially to cooperative settings, suggesting that this abstraction captures much of the structure needed for coordination. At the same time, explicit multi-robot training further improves performance, showing that shared-object coupling introduces coordination dynamics that are beneficial to learn directly. We validate the approach in simulation across varying team sizes and object geometries, and demonstrate sim-to-real transfer on hardware, where the learned controllers enable real humanoids to perform cooperative manipulation tasks.