Researchers led by Ting-Hsuan Chen have built Heave-Ho, a decentralised system that lets humanoid robots work together on long tasks that involve moving and handling objects, according to a paper on arXiv.

Each robot controls its own body from local observations, using a humanoid motion tracking model adapted for robots that push and pull on each other. A vision-language planner sets the steps, and a leader robot uses speech to keep the followers in step, without network communication.

Trained with two identical humanoids, the system scaled to teams of up to ten robots and worked with mixed teams without retraining. The team built a benchmark, HH-Bench, with 12 tasks for 2 to 10 humanoids, and the authors say real-world tests showed the robots working together on difficult tasks.

Source: arXiv Robotics (cs.RO)

Background

Ting-Hsuan Chen

Robotics researcher

Robotics researcher.