Boston Dynamics and the Toyota Research Institute (TRI) have demonstrated the use of a Large Behavior Model (LBM) to control Boston Dynamics’ Atlas humanoid robot in a recent research collaboration. The video released by the organizations shows Atlas carrying out a continuous sequence of tasks that combine object manipulation with locomotion, including packing, sorting, and organizing.
During the demonstration, the robot performed whole-body movements such as walking, crouching, and lifting, while responding to unexpected changes introduced mid-task. Unlike earlier methods, which often separate locomotion and manipulation, the project applied a single LBM to control the robot’s entire body, treating hands and feet in a unified way.
The work stems from a research partnership established in October 2024 between Boston Dynamics and TRI, designed to combine expertise in robotics hardware and artificial intelligence. According to the organizations, LBMs make it possible to introduce new robot capabilities through demonstrations rather than programming, which can accelerate the addition of new skills.
Scott Kuindersma, vice president of robotics research at Boston Dynamics, said the approach shows how training a single neural network to perform long-horizon manipulation tasks could support better generalization. Russ Tedrake, senior vice president of Large Behavior Models at TRI, said LBMs provide a new way to enable humanoids to carry out a wide variety of tasks in existing environments, with fewer demonstrations required as the models improve.
The project, co-led by Kuindersma and Tedrake, is aimed at advancing research on large models for whole-body robotic control, including manipulation and dynamic behaviors.
