Boston Dynamics Fitted Atlas With Wheels to Test Motor Limits

In a technical essay published by Boston Dynamics, controls engineer Xiaoyu Andy Wu reported that the robotics firm has fitted its Atlas humanoid robot with wheels instead of feet to test motor limits.

Wu stated that the engineering conversion required just a single custom adapter to mount an off-the-shelf scooter rim directly onto the robot’s existing ankle actuator. Boston Dynamics reported that no other major hardware changes were necessary because Atlas was designed from the start for high-volume production and simple servicing.

According to Boston Dynamics, the software modification was similarly straightforward. Engineers modeled each wheel as an ankle capable of continuous rotation, allowing them to swap the simulated feet out within the baseline robot model.

Atlas on a Roll | Classical Motor Controls and Whole-Body RL | Boston Dynamics

Wu explained that the wheels were not intended as a commercial product but served as a research testbed to force motors to deliver high torque and high speed continuously. Boston Dynamics reported that every launch, turn, and stop pushed the actuators across their full torque-speed range under real loads.

Interesting Engineering reported that the experimental setup provides a demanding environment for testing how the humanoid handles sustained high-speed movement and heavy loads. Wu stated that driving the robot with wheels and seeing it slide around resembled playing a video game.

Wu noted that the training pipeline successfully utilized the wheels during side dashes, automatically figuring out how to roll rather than drag across the ground. Boston Dynamics reported that Atlas also executed a 360-degree jump by running a trajectory-tracking training pipeline over a motion capture library of a human performing the maneuver.

According to Interesting Engineering, the robot rolled its lower body into position to prepare for takeoff before executing the jump on its first attempt. Wu stated that closing the sim-to-real gap required tuning low-level control parameters and adjusting simulated motor models to match physical measurements.

Photo of author

James Carter Senior News Editor

Senior Editor, News James is an award-winning investigative reporter known for real-time coverage of global events. His leadership ensures Archyde.com’s news desk is fast, reliable, and always committed to the truth.

PJ.2.1 Variant Spread Highlights Lingering Covid-19 Risks

Why the Midterm ‘Blue Wave’ Missed South Dakota