Boston Dynamics makes incredible robots, but we’re built on the human element: embracing challenges, getting creative, and having fun. With the FIFA World Cup 2026™ coming up, we were excited to work with Hyundai to try something new. The School of Football is a part of Hyundai’s Next Starts Now campaign, showing Atlas learning football and mastering a complex Ghost Rabona kick. The FIFA World Cup™ is an amazing stage to celebrate the ways that football brings people together, inspires them, and elevates new possibilities. We wanted to bring that energy into our lab and see if football can teach a robot how to move.

We see human athletes perform amazing acts of physicality every day. That’s a bar that we want to achieve for our robots as well. In robotics, we talk a lot about using data pipelines and flywheels to train generalist behaviors for robots. But, in some ways, it’s not that different from how people learn: human athletes improve by practicing, watching back game tape, analyzing their performance, and adjusting their training routines. We applied the same logic to our Atlas® humanoid robot, using human reference and reinforcement learning to train Atlas to perform a previously unseen trick shot.

The Ghost Rabona

Our goal was to showcase the incredible mobility, agility, and whole-body control enabled by our very capable humanoid robot. We came up with the Ghost Rabona, combining a fake out step over with a crossed-leg rabona kick. 

  • Step 1: Atlas walks towards the ball. 
  • Step 2: Atlas fakes the shot with its left leg. 
  • Step 3: Atlas crosses its right leg behind to strike the ball.

This move is complicated for a human to perform, much less a robot. Atlas needs to move fast for a convincing fake. It needs power and agility to fully take off from the ground and land again, while still remaining balanced to complete the kick. This combination of different skills pushes the limits of physical intelligence. Making a robot move dynamically and powerfully is exactly the kind of research problem we love to solve at Boston Dynamics.

Learning from (Human) Demonstrations

Just as a human athlete must practice to master intricate movements, Atlas requires training to perfect its skills. We use reinforcement learning to teach the robot. Atlas undergoes extensive training in simulation to learn how to handle the physics of its own body to mimic human motion.

The first step is to create a reference input from human demonstration. Human demonstration simplifies robot programming, turning what used to be a complex coding process into an intuitive task. These demonstrations can come from various sources—including video, teleoperation, and motion capture. For this particular task, we utilized an optical motion capture system to record high-fidelity kinematic data of a person performing the action.

For the Ghost Rabona, we worked with a football player to develop and record a super dynamic motion. I (Roberto) also took a turn in the motion capture suit for some of the drills and basic kicks, so a lot of the motions that you see is me performing those motions and then training a robot to replicate my behaviors.


Using human motion capture allows for the rapid collection of numerous demonstrations while embedding them with a distinctively “human” style. The difficulty lies in adapting these movements to Atlas’ kinematics, as there isn’t a direct correspondence between human and robot morphology. Despite Atlas’s humanoid appearance, its kinematic structure is fundamentally unique.Through a retargeting process, we map human motion into the Atlas kinematics. 

Once we have that reference trajectory, the machine learning process begins. We use  reinforcement learning to train a policy for that behavior. The robot still has to figure out the physics of how to move its body. It’s not just the motion, but it’s actually the actuation, how it controls the motors in order to balance and imitate the motion performed by the human.

Atlas practices in simulation, with thousands of simulations in parallel happening in GPUs in the cloud. And over time the system learns the actions necessary to realize the behavior in simulation. It’s similar to how humans learn through trial and error, except the robot gets to try things much faster than a human could: the equivalent of a full year’s worth of physical trial and error in the span of just 24 hours.

Once Atlas learns to perform the skill in simulation we deploy on the hardware. For almost all of Atlas’ skills, the learned policy worked first try on the robot. If something fails we go back to our training, make adjustments, and improve the behavior.

Transferrable Skills

Every time Atlas moves into a new environment, there’s going to be new training—similar to any new hire learning the ins and outs of this specific job—where it can operate, what it needs to do, how it needs to interact with the world. While the task is not the same, many of the same tools we use to train the robot for football transfer to training the robot to do a job in a warehouse or in a factory.

There’s also transferable skills we can hone playing football that translate into useful tasks. Unlike other sports that separate locomotion from manipulation, football demands mastering both at once. Athletes must balance and run while precisely controlling a ball with their feet—a significant hurdle for humanoid robots. Success requires a whole-body controller that dynamically coordinates every joint as a single system, allowing the robot to strike the ball powerfully while maintaining balance. This same whole-body coordination is the essence of what Atlas needs to perform manipulation tasks in the workplace.

More broadly, training generalist behavior for robots requires a vast variety of data and training. We also train Atlas on more mundane, practical applications that it will do in real work environments, but thinking—and training—outside of the box is one of the keys to unlocking generalizable behaviors.

While Atlas was engineered for industrial and factory applications, its capabilities extend far beyond those environments. The demonstrations featured here illustrate the extraordinary power and agility that define it. These skills are just the beginning of what Atlas can do. We’re excited to be showing off our skills for the FIFA World Cup™ and we’re excited to show the Next of robotics. Next starts now.