science
KAIST Humanoid Robot Demonstrates Running, Moonwalking and Ball Kicking in Field Test
Engineers at Korea Advanced Institute of Science and Technology developed a humanoid robot that sprints, moonwalks and kicks a soccer ball on a field, using custom motors and AI for controlled movements. The robot, standing five feet five inches tall and weighing 165 pounds, achieves speeds up to 7.3 miles per hour and climbs steps over one foot high.
Apr 5, 7:55 AM·2m read1 source
Substrate placeholder — needs reviewAudio version
Tap play to generate a narrated version.
Engineers at the Korea Advanced Institute of Science and Technology (KAIST) have developed a humanoid robot capable of sprinting, moonwalking and kicking a soccer ball during a recent field test on a soccer field. The robot performed these actions with smooth, repeatable movements powered by custom-designed motors and artificial intelligence.
Led by researcher Hae-Won Park, the team constructed the robot from scratch, avoiding off-the-shelf components to optimize performance. This approach allows for precise control over power distribution, resulting in enhanced torque and real-time responsiveness.
The robot measures five feet five inches tall and weighs 165 pounds, enabling it to maintain balance during fast movements.
A key component is the Quasi-Direct Drive system, which combines strong motors with low gear ratios for quick responses and stability. This setup, paired with a compact gearbox, reduces weight and improves efficiency. In tests, the robot reached speeds of up to 7.3 miles per hour and climbed steps taller than one foot.
The team continues to refine these capabilities for higher performance levels. These features were demonstrated in a non-laboratory environment, highlighting the robot's ability to execute actions consistently without errors.
The robot's movements appear natural due to Physical AI integration, which uses deep reinforcement learning trained on human motion data. Initial training occurred in simulations before transfer to the physical robot, enabling fluid transitions between actions like running, jumping and direction changes.
For instance, during the field test, the robot sprinted across the soccer field, kicked a ball toward the goal and adjusted direction seamlessly. This learning method contrasts with pre-programmed sequences, allowing adaptations that mimic human-like control.
