VioLA: Learning Generalist Humanoid Control Policies from Human Data
VioLA turns human motion recordings into a usable action space for humanoid control.
The paper says the policy predicts body and hand motion latents, which pretrained controllers execute on the robot. That lets human and robot demonstrations share the same latent space, expanding training to 140.6 million frames, mostly human. In reported real-robot tests, VioLA reaches 100% zero-shot locomotion success and 88.6% manipulation success without task-specific fine-tuning. Code and checkpoints are planned for release. ArXiv · AI/CL/LG's note
The paper says the policy predicts body and hand motion latents, which pretrained controllers execute on the robot. That lets human and robot demonstrations share the same latent space, expanding training to 140.6 million frames, mostly human. In reported real-robot tests, VioLA reaches 100% zero-shot locomotion success and 88.6% manipulation success without task-specific fine-tuning. Code and checkpoints are planned for release. ArXiv · AI/CL/LG's note
score 5