Researchers led by Allen Devaraj have shown a humanoid robot that leans on and braces against its surroundings to reach objects far beyond its normal reach, according to a paper on arXiv under review for ICRA 2027. Most whole-body control for humanoids tries to avoid or accommodate contact, but reaching objects on far-away surfaces requires it, the authors write.

Their sampling-based model predictive controller handles leaning, bracing, reaching and recovery on a Unitree H1-2 humanoid using only its own sensors and fiducial markers. It runs in real time at 34 Hz on a laptop, transferred from simulation to the real robot, extended its reach by about 50 cm and held the loaded posture for more than four minutes before its pose estimates drifted.

Across 54 trials with six targets and three table heights, the robot reached the commanded target in 89% of trials and completed the whole reach-and-recovery cycle in 81%.

Source: arXiv Robotics (cs.RO)

Background

Allen Devaraj

Robotics researcher

Robotics researcher.

Unitree H1-2

Humanoid robot

A humanoid robot made by Unitree Robotics.

Unitree Robotics

Robot maker

Unitree Robotics is a Hangzhou company that makes four-legged robot dogs and humanoid robots, including the G1, which research teams at Apple and other labs use.