Field test video depicts South Korea’s KAIST Humanoid v0.7 robot’s outstanding moonwalk and soccer skills, highlighting its precise high-speed locomotion.
Weighing 165 pounds (75 kilograms), the humanoid was developed at the Korea Advanced Institute of Science and Technology (KAIST)’s Dynamic Robot Control Design Laboratory (DRCD Lab) under the direction of Dr. Hyewon Park.
The clip highlights humanoid use of physical AI, an approach that allows autonomous machines to recognize, interpret, and perform complex actions in real-world environments. They can be seen running, jumping, shooting on goal, and performing fluid dance moves on the soccer field.
The 5-foot-5-inch robot combines advanced hardware and an intelligent control system, the researchers said. Designed to be reliable, efficient, and scalable.
Physics AI behavior
Rather than relying on off-the-shelf parts, Korean researchers independently developed all components of v0.7, including the motor, gearbox, and motor driver, making it technically independent.
This makes it possible to optimize torque density and power-to-weight ratio. According to the team, both are important for high-speed locomotion and dynamic balance.
At the same time, the actuation system of the robot is based on a quasi-direct drive (QDD) architecture. It combines a high torque motor with a low gear ratio for more precise control while increasing responsiveness.
This is supported by a custom-designed 3K compound planetary gearbox, which provides high gear reduction in a compact single-stage configuration. The result is a lighter, more efficient system that can handle demanding tasks such as running, jumping, and rapid direction changes.
Hae-Won Park emphasized that the robot can run at speeds of up to 10.7 feet per second (about 7.3 miles per hour or 12 kilometers per hour) on flat ground. It can climb steps over 12 inches (30 centimeters) in height.
Professor Park said, “The team will further improve its performance, aiming for a running speed of 4.0 meters/second (approximately 14 km/hour), the ability to climb ladders, and climb steps over 40 centimeters.”
Additionally, the humanoid knee actuator can deliver peak torque of up to 320 Newton meters (Nm). Meanwhile, the ankle actuator is optimized for fast response and stability. This allows the system to perform complex movements that are both powerful and delicate.
powerful software
To create smoother, more natural movements, researchers integrated deep reinforcement learning (DRL) with human movement data. They trained the system in simulation and used human movements as pre-actions. This allows v0.7 to avoid the jerkiness often seen in purely AI-driven systems.
It also incorporates motor operating range (MOR) modeling to constrain the simulation to match the physical limitations of the hardware, using a hybrid approach known as modular residual learning.
“This achievement is an important milestone in achieving independence for both the hardware and software aspects of humanoid research by protecting core components and AI controllers with proprietary technology,” Park said.
The robot can also navigate uneven terrain using only proprioception, without relying on visual sensors. This is particularly relevant in industrial environments where visibility may be limited.
Park concluded in a press release, “In order to solve the complex demands of actual industrial sites, we will further develop it into a complete humanoid, including the upper body, and develop it into a next-generation robot that can collaborate with humans.”
Meanwhile, DRCD Lab is developing DynaFlow, a framework aimed at enabling robots to learn complex tasks directly from human demonstrations. This could allow humanoids to perform practical tasks, from handling tools to operating machines.
