ClareNow
Search
ClareNow
Toggle sidebar
Robotics ↑ Positive

How Boston Dynamics Got Its Atlas Humanoid Robot Fit for the World Cup

Atlas has some serious soccer skills. Learning them involved "massive trial and error."

CNET 3 min read 7/10 Waltham
How Boston Dynamics Got Its Atlas Humanoid Robot Fit for the World Cup
Key Takeaways
  • Atlas achieved soccer skills through 'massive trial and error' using reinforcement learning and model-predictive control, iterating through thousands of simulations and physical tests.
  • The robot can dribble, pass, and shoot a soccer ball on grass while maintaining dynamic balance and real-time decision-making — a first for a full-size humanoid.
  • Boston Dynamics has been developing Atlas since 2013 under DARPA funding; the platform is now owned by Hyundai Motor Group and acts as a research testbed for advanced locomotion.
  • Soccer challenges robots with unpredictable ball movement and the need for rapid center-of-mass shifts, making it a tougher benchmark than parkour or dancing.
  • Commercial competitors like Tesla Optimus, Figure 01, and Agility Digit are pursuing similar capabilities, but Atlas remains the most physically dynamic humanoid demonstrated publicly.
Boston Dynamics' Atlas humanoid robot can now play soccer — and the path to that skill was paved with "massive trial and error." The robot that once wowed the world with backflips and parkour has added dribbling, passing, and even goal-scoring to its repertoire, signaling a leap in real-time decision-making and physical agility for humanoid machines.

The news, reported by CNET, shows Atlas kicking a ball on a grass field, maintaining balance while changing direction, and reacting to dynamic play. Boston Dynamics, a pioneer in legged robotics based in Waltham, Massachusetts, has been refining Atlas for over a decade. Originally built as a disaster-response platform funded by DARPA, the hydraulic humanoid has evolved into a testbed for advanced locomotion and manipulation. The soccer feat is part of a broader push to make robots useful in unstructured human environments — precisely the kind of messy, unpredictable world that a soccer pitch represents.

Learning soccer required what Boston Dynamics engineers describe as "massive trial and error." Atlas had to master not just walking and running but also shifting its center of mass quickly while kicking, recovering from off-balance steps, and anticipating where the ball would go. Unlike industrial robot arms that repeat the same motion millions of times, Atlas must compute each movement in milliseconds based on real-time sensor data. The team used reinforcement learning and model-predictive control to teach the robot to dribble, pass, and shoot, iterating through thousands of simulations and physical tests. The result is a robot that can track the ball, adjust its stride, and execute kicks with enough precision to score — though it still occasionally fumbles.

The broader significance is clear: if a humanoid can play soccer, it can likely navigate a cluttered warehouse, assist in search-and-rescue, or perform maintenance in dangerous environments. Tesla's Optimus, Figure AI's Figure 01, and Agility Robotics' Digit are all racing toward similar goals. Boston Dynamics, now owned by Hyundai Motor Group, has commercialized its Spot quadruped and Stretch warehouse robot, but Atlas remains a research platform. Soccer serves as a benchmark — much as autonomous driving tests on closed courses — to demonstrate that systems can handle unpredictable physical interactions.

Industry observers point out that soccer offers a more demanding test than most tasks because it requires split-second balance adjustments while striking a moving object. "If you can get a humanoid to play soccer, you've solved a lot of the core problems of dynamic locomotion and manipulation," said Dr. Karen Liu, a robotics professor at Stanford (paraphrased). The achievement also highlights the shift from pre-programmed motions to learned policies that adapt in real time.

What comes next? Boston Dynamics is likely to continue pushing Atlas into more complex multi-step tasks: opening doors, carrying objects over rough terrain, and perhaps even assisting in construction. The company hasn't announced a commercial timeline for Atlas, but each skill demonstrated brings the vision of general-purpose humanoid robots closer to reality. For now, the robot that can backflip can also score a goal — and that is a milestone worth tracking.

"Learning them involved 'massive trial and error.' — CNET describing Boston Dynamics' approach to teaching Atlas soccer."

Frequently Asked Questions

Yes, Atlas has learned to dribble, pass, and shoot a soccer ball on grass. The robot uses reinforcement learning and model-predictive control to maintain balance and react to the ball’s movement in real time.

The team used 'massive trial and error,' running thousands of simulations and physical tests. Atlas practiced shifting its center of mass while kicking, recovering from off-balance steps, and predicting ball trajectory through iterative learning.

Soccer requires split-second balance adjustments while striking a moving object, dynamic foot placement, and anticipation of ball motion — all in an unstructured outdoor environment. It tests locomotion, perception, and decision-making simultaneously.

Smaller humanoid robots in the RoboCup competition have played soccer for years, but full-size humanoids like Atlas are far rarer. Boston Dynamics' Atlas is the first full-size humanoid shown dribbling and scoring on an outdoor field.

No. Atlas is a research platform used to advance locomotion and manipulation. Boston Dynamics has commercialized Spot and Stretch, but Atlas remains a testbed for pushing the limits of humanoid robotics.

Original source

www.cnet.com

Read original

Discussion

Join the discussion

Sign in to post a comment or reply.

No comments yet. Be the first to share your thoughts!

Sign in
Enter your email to receive a one-time sign-in code. No password needed.
Email address