MotionDisco humanoid robot learning
by Canyon Owen: artificial intelligence
The framework known as Motion Disco lets humanoid robots learn skills like climbing, balancing, and object handling without any human demonstrations or remote control. By combining AI with motion planning, the system allows these machines to discover new methods for task completion rather than relying on pre-set scripts. This approach enables the robots to identify multiple ways to finish a single job entirely on their own.
Transcript (en)
Researchers have taught humanoid robots to learn skills like climbing, balancing, and object handling on their own, using a new framework called Motion Disco. The robots can discover complex behaviors without human demonstrations or teleoperation. The system combines a large language model with motion planning to explore different interaction strategies, allowing robots to find multiple ways to complete a task.
