A machine and a learned controller
Boston Dynamics develops robot hardware and the control systems that make it useful. Its Atlas work combines whole-body motion, perception and manipulation, where reaching for an object changes balance and can require the robot to reposition its feet. The learning problem extends beyond generating a verbal task plan. [1]
Its August 2025 work with Toyota Research Institute demonstrated a language-conditioned policy performing manipulation and locomotion together. The technical account describes collecting teleoperated demonstrations, curating the data, training a neural policy and testing it across tasks. Teleoperation supplies training experience; the learned execution is a separate stage. [1][2]
Research Atlas and product Atlas
The January 2026 product announcement distinguished a new Atlas product version from the prototype performing onstage. It described manufacturing beginning and selected fleets scheduled for customers during 2026. The robot also supports autonomous, teleoperated and tablet control, so its presence at a site alone does not identify which mode performed a task. [3]
Those records show a path from research demonstrations toward industrial products. They do not establish that every learned behavior demonstrated on one Atlas generation is available on every customer robot, or that scheduled deployments have all been completed. [3][1]
Sources
- Large Behavior Models and Atlas Find New Footing
Boston Dynamics · retrieved Oct 23, 2026
- AI-Powered Robot by Boston Dynamics and Toyota Research Institute Takes a Key Step Towards General-Purpose Humanoids
Toyota Research Institute · Aug 20, 2025
- Boston Dynamics Unveils New Atlas Robot to Revolutionize Industry
Boston Dynamics · Jan 5, 2026