Event
RynnVLA-Latent Robot Manipulation Foundation Model Paper Released
1 reports · 1 sources
The research team released the RynnVLA-Latent robot manipulation foundation model framework, whose core is using RynnLAM to learn continuous, 3D-grounded latent actions from visual changes without requiring native robot control signals, thereby converting heterogeneous interaction videos into scalable policy pretraining supervision. The team states that it has converted 30.1K hours of human, robot, and simulation videos into 745.52M latent action descriptions for policy pretraining; on LIBERO-Plus, it reached 84.7% without benchmark-specific adaptation, and it was tested on six real manipulation tasks across dexterous hand and mobile gripper platforms.
Report timeline
- RynnVLA-Latent: Scaling Robotic Manipulation Foundation Models with 3D-Grounded Latent Actions
The research team proposes the RynnVLA-Latent framework, which uses RynnLAM to learn continuous, 3D-grounded latent actions from visual changes without native robot control signals, converting 30.1K hours of human, robot, and simulation video into 745.52M latent action descriptors for policy pretraining.
arXiv cs.RO ·