Atlas Humanoid Gains LBM Skills From Boston Dynamics and TRI
Boston Dynamics and the Toyota Research Institute announced a significant advance in robotics and artificial intelligence, demonstrating a Large Behaviour Model (LBM) powering the Atlas humanoid robot. In a jointly released video, Atlas performs a long, continuous sequence of complex tasks that combine object manipulation with locomotion. By adopting LBMs, engineers can add new capabilities quickly, without writing a single new line of code, replacing labour-intensive hand-programming.
The video shows the humanoid using whole-body movements such as walking, crouching and lifting to complete packing, sorting, and organising tasks. Researchers introduce unexpected physical challenges mid-task, including closing the lid of a box and sliding it across the floor, prompting Atlas to self-adjust in real time. Previous humanoids that showed similar skills typically separated low-level walking and balancing from arm control for manipulation; in this project, a single Large Behaviour Model directs the entire robot, treating the hands and feet almost identically for unified, whole-body control.
The October 2024 joint research partnership between Boston Dynamics and TRI delivered this breakthrough by combining their strengths to accelerate innovative robot development. The result reinforces the strong potential of AI technologies in creating general-purpose humanoid assistants capable of dynamic behaviours and precise manipulation.
“This work provides a glimpse into how we’re thinking about building general-purpose robots that will transform how we live and work,” said Scott Kuindersma, vice president of Robotics Research at Boston Dynamics. “Training a single neural network to perform many long-horizon manipulation tasks will lead to generalisation, and competent robots like Atlas present the fewest barriers to data collection for tasks requiring whole-body precision, dexterity, and strength.”
“One of the main value propositions of humanoids is that they can achieve a huge variety of tasks directly in existing environments, but the previous approaches to programming these tasks simply could not scale to meet this challenge,” said Russ Tedrake, senior vice president of Large Behaviour Models at Toyota Research Institute. “Large Behaviour Models address this opportunity in a fundamentally new way – skills are added quickly via demonstrations from humans, and as the LBMs get stronger, they require fewer and fewer demonstrations to achieve more and more robust behaviours.”
Co-led by Scott Kuindersma and Russ Tedrake, the project investigates fundamental questions about humanoid robots and Large Behaviour Models, advancing understanding of large models for whole-body control, including advanced manipulation and dynamic behaviours that push the state of the art in robotics and AI.