AthenaZero Brings Two-Handed Speed to Robotics, Juggles with Ease

Footage from the Robotics and AI Institute shows AthenaZero, a robot, juggling three balls with only its bare hands. There are no external cameras whirling around it, and no extra tools assisting in guiding the balls into place; only the robot monitoring, adjusting, and maintaining the rhythm as it learns from each bounce off its palms with every contact.

The institute’s engineers created AthenaZero as a prototype for activities that demand lightning-fast reactions from both arms functioning in unison. Standing 5 feet 3 inches tall, with a six-foot wingspan. AthenaZero has a one-degree-of-freedom torso, two seven-degree-of-freedom arms, and several six-degree-of-freedom hands. The entire system relies on 20 strange actuators that power 27 joints, and it can easily manage more than 6.5 pounds, but what’s even more surprising is how light it moves.

The key to this was the motor design they used, which included quasi-direct drive motors that sit very near to the torso and step via a low gear ratio system with over 97% efficiency. This keeps the weight at the wrist to around 9 pounds, which is roughly the same amount of weight we experience in our own arm. Shifting some of the mass towards the body allows the robot to move without exerting a great deal of effort and eliminates the sharp jolts of stopping or swinging through a tight turn. Plus, the motor current readings provide real-time torque feedback, allowing the robot to sense what is going on without the need for additional sensors. This all adds up to a system that is extremely compliant and easy to nudge out of the way, which allows it to absorb minor errors during quick motion rather than struggle against them.

All of this is made possible by some extremely precise control methods that match the stiffness of a human arm based on its phase of motion. The trajectory planning looks at the big picture, while the impedance adjustments fine-tune how hard or soft the hands give or resist when they make contact with something, and reinforcement learning helps the robot improve its timing through repeated trials, especially when things get a little unpredictable and the objects begin to slip or bounce around.

Older robot designs frequently relied on high gear ratios, which added a lot of weight and made each movement stiff. This made quick direction changes extremely difficult, and anything unexpected might easily throw something off balance or cause damage to the parts or the object itself. AthenaZero, on the other hand, completely avoids this by keeping mass extremely low and enabling natural movement to occur. The arms remain extremely responsive, able to transition from lifting something large to gently guiding something fragile in an instance.