Event
MindOn releases Psi R0.5 end-to-end VLA model with lower data requirements
Evidence notes
- MindOn announced Psi R0.5 as an enhanced end-to-end VLA model built on a reinforcement-learning-augmented hierarchical architecture.
- The company said the release improved generalization, dexterity, chain-of-thought behavior, and long-horizon task capability while cutting data requirements for generalized grasping training.
- The announcement also tied the launch to four disclosed papers covering grasp generalization, object retrieval in stacking scenarios, environment-assisted grasping, and VLA safety alignment.
Company context
MindOn is a Shenzhen embodied-AI company focused on the software and algorithm layer for humanoid robots rather than its own robot chassis. Public reporting describes the company as building control and model systems that can run on existing humanoid hardware, with its best-known demonstrations shown on Unitree platforms.
Context
- Company
- MindOn
- Segment
- Foundation Models
- Event type
- Product Launch
- Added to Korthos
- Apr 2, 2026
- Geography
- Shenzhen, Guangdong · China