TECH – China’s Unitree Robotics has pushed the boundaries of humanoid mobility by sending its compact G1 humanoid robot on a fully autonomous walk across a frozen snowfield in the Altay region of Xinjiang—one of the coldest environments ever attempted by a bipedal robot. Operating in temperatures plunging to –47.4°C (–53°F), the G1 completed more than 130,000 consecutive steps, demonstrating a level of balance, resilience, and autonomy rarely seen outside controlled laboratory settings.
Standing 127 centimeters tall and weighing around 35 kilograms, the Unitree G1 is built with a lightweight but robust frame designed for dynamic movement. During the test, the robot traversed icy, uneven terrain while maintaining stable posture and consistent gait—conditions that would easily destabilize most humanoid platforms. Its movement was not guided remotely; instead, the robot relied entirely on onboard intelligence to perceive the ground, adjust foot placement, and recover balance in real time.
The robot’s perception system combines LiDAR, depth cameras, and inertial sensors, allowing it to build an internal understanding of its surroundings even when visual cues are distorted by snow glare and low contrast. Navigation was assisted by China’s Beidou satellite positioning system, enabling the G1 to follow a precise route and trace a large Winter Olympics emblem across the snow—an area measuring roughly 186 meters long and 100 meters wide.
Read More: China Cultivates Military Rubber in Gobi Desert to Cut Imports
To withstand the extreme cold, engineers equipped the G1 with practical adaptations rather than redesigning its core structure. An insulated outer jacket protected internal electronics, while custom plastic shielding around its legs reduced exposure of joints, actuators, and battery systems to ice and wind. These additions highlight how modular protection can significantly expand a robot’s operational environment.
Powering the robot is a 9,000 mAh quick-swap battery, paired with Unitree’s proprietary UnifoLM control system, which uses reinforcement-learning-based motion algorithms. This allows the G1 to dynamically adjust stride length, joint torque, and center of gravity—capabilities that are critical when walking on slippery or unpredictable surfaces. Under normal conditions, the robot can walk at speeds approaching 2 meters per second.
The experiment underscores a growing shift in humanoid robotics: from machines designed primarily for indoor demos to platforms capable of functioning in harsh, real-world environments. With its successful snow walk, the Unitree G1 offers a glimpse into a future where humanoid robots may assist in search-and-rescue missions, remote inspections, and operations in extreme climates, extending human reach into places that are dangerous or inaccessible.
