Dyna Robotics Introduces Dyna-2: A World-Action Model Pre-Trained on 1 Million Hours of Human Video
Dyna Robotics has launched Dyna-2, a new world-action model trained on over a million hours of human egocentric video. This model demonstrates improved generalization to robot data and cross-embodiment capabilities.
- Dyna-2 is a world-action model trained on over 1 million hours of human video.
- The model demonstrates successful transfer learning from human video data to robot data.
- Video co-training enhances Dyna-2's ability to generalize actions across different embodiments.
Dyna Robotics has unveiled Dyna-2, a significant advancement in world-action modeling. The model leverages an unprecedented dataset of over one million hours of egocentric human video for pre-training.
Key findings from their technical report include the establishment of a scaling law specifically for human data, demonstrating how performance improves with increased training hours up to the million-hour mark. Furthermore, Dyna-2 shows the first successful transfer of this scaling law to robot-specific data, indicating its potential for real-world robotic applications.
The research also provides evidence that co-training with video data significantly drives cross-embodiment generalization, meaning the model can better apply learned actions across different physical forms or environments.
Provides a new model architecture and training methodology for robotics and embodied AI.
Opens possibilities for more adaptable and generalizable robots in various industries.
Highlights advancements in embodied AI, a growing area of interest.
Represents progress in AI's ability to learn complex actions from real-world human behavior.
- world-action model
- An AI model designed to understand and predict actions within a physical environment.
- egocentric video
- Video footage recorded from the perspective of a moving person or agent.
- cross-embodiment generalization
- The ability of an AI model to apply learned skills or knowledge across different physical forms or robotic platforms.
Surgical WAM: A World-Action Model for Data-Efficient Surgical Robot Learning
Explainer: What is Unitree and why are China’s humanoid robot makers racing to list? - Reuters
RoboticsThe first self-driving vehicle on Mars has proven to be a smashing success
Air Force expands autonomous flight tests with live, AI-enabled intercepts - DefenseScoop
RoboticsUS company’s AI lets Ukraine’s cheap kamikaze drones track targets on their own
Beyond Digital as Usual, Artificial Intelligence for Accessible Learning - UNICEF
UNICEF is exploring the use of artificial intelligence to improve access to learning for children worldwide.
China urged to avoid ‘us or them’ split with US over AI governance - South China Morning Post
China is urged to avoid a 'us or them' split with the US over AI governance, amid growing tensions between the two nations.
Development of artificial intelligence could surpass China’s industrial boom, say Capital Group - The Corner .eu
Capital Group predicts AI development could exceed China's industrial boom, The Corner reports. This forecast highlights the potential for AI to drive significant economic growth.
HardwareOrganic-looking brake assemblies debut on new Czinger 21C Spyder
Czinger’s 21C Spyder hypercar features brake assemblies produced using topological optimization and additive manufacturing, marking a first for automotive braking systems.
Naabik'iyati' Committee approves the establishment of the Navajo Nation Artificial Intelligence Policy Working Group WINDOW ROCK, Ariz. – The Naabik'iyati' Committee approved a position statement and the creation of a Navajo Nation Artificial Intelligence P - facebook.com
The Naabik'iyati' Committee has approved the creation of a Navajo Nation Artificial Intelligence Policy Working Group. This group will develop a position statement on AI policy for the Navajo Nation.
AI ToolsMeet Needle 2: An Open 45M-Parameter Tool-Calling Model That Ships as a 14MB Binary and Runs a Full Session in 28MB of RAM
Cactus Compute has released Needle 2, a 45M-parameter AI model that runs efficiently on low-end hardware.