ShengShu Technology's Motus2 Achieves 84% Success in Robotic Dexterity Tasks
Event summary
- ShengShu Technology unveiled Motus2, a self-evolving general world model for robotic dexterous manipulation, at the 2026 Inclusion Conference on the Bund on September 10, 2026.
- Motus2 achieved an average success rate of 84% across five primary real-robot tasks, with a 10 percentage point improvement in success rate through model-based reinforcement learning.
- The model integrates action generation, consequence prediction, and outcome evaluation within a single video-action model with shared parameters.
- Motus2 draws on 130,000 hours of human manipulation data and over 100 hours of robot trajectories for training.
The big picture
ShengShu Technology's Motus2 represents a significant advancement in robotic dexterous manipulation, bridging the gap between predictive modeling and real-world action. The model's success rate improvements highlight the potential for AI-driven robotics to handle complex tasks, positioning ShengShu as a key player in the development of autonomous world agents. The integration of human and robot data underscores the growing importance of hybrid training approaches in AI development.
What we're watching
- Technical Scalability
- Whether Motus2 can sustain performance improvements across a broader range of real-world tasks and environments.
- Industry Adoption
- The pace at which robotic dexterity solutions like Motus2 are integrated into industrial and consumer applications.
- Competitive Positioning
- How ShengShu Technology differentiates Motus2 in a market increasingly focused on autonomous robotic systems.
Related topics