From Moonwalking to Heavy Lifting: KAIST Humanoid v0.9 Moves 35 kg Objects

KAIST’s latest humanoid update puts its strength to work: Humanoid v0.9 pushes objects weighing up to 35 kg and lifts objects up to 20 kg, according to a new demonstration released on October 19. The robot uses onboard visual feedback as it handles objects to arrange the letters “KAIST.”

Key Takeaways

Show all

• KAIST Humanoid v0.9 demonstrates pushing objects up to 35 kg and lifting objects up to 20 kg.
• DRCD Lab developed the humanoid in collaboration with RAILAB, which developed its arms.
• The update combines onboard visual feedback with interaction skills developed through deep reinforcement learning and learning from human demonstrations.

From the Moonwalk to moving loads

The new video, titled Whole-Body Locomanipulation with KAIST Humanoid v0.9, shifts attention toward coordinating movement and object handling. KAIST credits its Dynamic Robot Control & Design Laboratory, or DRCD Lab, and arm-development collaborator RAILAB.

It builds on the platform we covered in KAIST Humanoid v0.7: High-Speed Locomotion and the “Moonwalk” of Technical Independence. That update emphasized agile locomotion and the team’s development of core hardware in-house. The lab’s platform page describes the humanoid as a full-size robot for dynamic locomotion with a custom actuator system.

The new demonstration brings manipulation into that picture. Whole-body locomanipulation means coordinating balance, body movement and contact with objects: applying force through the arms also changes the demands on the legs and torso.

Two different measures of strength

The pushing and lifting figures describe different tasks. Moving a 35 kg object along a surface does not require supporting its entire weight, and the resistance depends on the contact conditions. The 20 kg lifting demonstration therefore provides a separate measure of the robot’s handling capability.

KAIST says the interaction skills were developed through deep reinforcement learning and human demonstration-based learning. The video description does not detail how those methods are combined or how tasks are commanded, so onboard visual feedback alone should not be read as evidence of unrestricted autonomy.

For now, the update shows the research platform extending from agile movement into visually guided physical work. It does not establish a commercial payload rating or sustained performance across repeated industrial tasks.