Atlas humanoid robot gets AI "power": Boston Dynamics teams up with Toyota Research Institute for AI behavior system application

Author: Li Dan

Source: Hard AI

Humanoid robot development is expected to achieve new breakthroughs, with the dancing robot Atlas set to be enhanced by advanced artificial intelligence (AI) systems.

On Wednesday, October 16th, Eastern Time, Boston Dynamics, a robot technology company under Hyundai Motor, and Toyota Motor jointly announced at Toyota Research Institute (TRI) in the United States that the two institutions in the AI and robotics fields will collaborate in Boston to accelerate the development of general humanoid robots using TRI's Large Behavior Model (LBM) and Boston Dynamics' Atlas robot.

The Large Behavior Model, or LBM, is a complex AI system. By learning from a large dataset of human interaction data, it can understand and generate human-like behaviors, replicating the subtle differences in human communication and decision-making. LBM is typically used in the field of robotics and human-machine interaction to make robot interactions with humans more natural, understand human gestures, and respond appropriately.

The collaboration with TRI means that Boston Dynamics will further advance the application of advanced AI technology in Atlas. In April of this year, the company introduced an electric version of Atlas and announced the development of a commercial version that will eventually be sold to the manufacturing industry. Boston Dynamics stated that the electric Atlas can lift objects heavier than elite human athletes, and a video released by the company shows that it can move in ways that humans cannot, such as bending its knees backward and rotating its torso 360 degrees.

The joint announcement on Wednesday stated that the latest generation of Atlas is the result of years of collaborative design between Boston Dynamics' hardware and software, aiming to build the most powerful humanoid robot platform in terms of both physical capabilities and software for writing full-body behaviors. Therefore, Atlas has become an ideal platform for advancing the scientific skills of AI manipulation.

TRI is currently a leader in the development of robot LBMs. TRI's pioneering work in diffusion strategies has successfully applied generative AI to the agile control of robots. TRI has also played a leading role in the development of open-source robot AI models and datasets. Leveraging additional advantages in computer vision and Large Language Model (LLM) training, TRI's LBM development aims to achieve a nimble control basic model that combines multitasking, vision, and language conditions.

Robert Playter, CEO of Boston Dynamics, highly praised the collaboration with TRI, stating that "the robotics industry has never been more exciting." This collaboration is a model that will demonstrate how two companies with strong research foundations can work together to address many complex challenges and create useful robots that solve real-world problems.

Gill Pratt, Chief Scientist of Toyota and CEO of TRI, stated that recent advances in AI and machine learning have enormous potential to advance physical intelligence. Adopting TRI's most advanced AI technology on Boston Dynamics' hardware represents a "game-changing" opportunity for both institutions TechCrunch commented that in the field of humanoid robots, Boston Dynamics' main competitors are Agility, Figure, and Tesla, who mainly choose to build AI teams internally. In contrast, the collaboration between Boston Dynamics and TRI is particularly interesting because the two institutions belong to Hyundai and Toyota respectively, which are old rivals in the automotive industry