Dyna Robotics has unveiled Dyna-2.1, a physical robot agent built upon their Dyna-2 world-action model. This model was trained on one million hours of human video and demonstrates impressive long-horizon autonomy, achieving a 99.4% success rate in folding restaurant napkins for 24 hours. Dyna Robotics is focusing on real-world production by enabling robots to learn new tasks from as little as 45 minutes of demonstration, aiming to overcome data scaling challenges in robotics. AI
IMPACT Sets new SOTA for physical robot autonomy and task learning from limited demonstrations.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- AlphaGo
- Din Tai Fung
- Dyna-2
- Dyna-2.1
- Dyna Robotics
- Google DeepMind
- GPT-6
- Nvidia
- Swyx
- University of Pennsylvania
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →