Robot Era, a Chinese humanoid robot developer, became the first company to put a humanoid robot through a walking test on the Great Wall of China.
In a video released yesterday by Robot Era, XBot-L, a full-sized humanoid prototype standing 1.65 meters tall, is seen walking along sections of the Great Wall, waving hello, practicing shadowboxing and performing other movements.
Robot Era took the machine out to showcase its locomotion, dexterity and self-balancing capabilities, in particular, its adaptability to a complex outdoor environment on the Great Wall.
This ancient fortress, with its weathered stone walls and pavements, poses a big challenge to any bipedal humanoid robot.
Normally, wheeled robots or humanoid models without intensified kinematic support are prone to fall when walking on uneven surfaces or stepping over potholes on an endless flight of stairs.
Besides, inside dimly lit archways within the guard towers, a robot can easily lose directions and hit the walls for want of reliable perception and navigation.
Thanks to advanced perceptive reinforcement learning (RL) algorithms, XBot-L can sense the environment in front of it, maintain its balance and adjust its pace and gait accordingly.
Perceptive and decision-making capacity
“Perceptive RL algorithms help to strengthen the robot’s perceptive and decision-making capacity in the face of unfamiliar terrains,” Yue Xi, co-founder of Robot Era, says. “The robot thus can recognize complex road conditions and adjust its walking stance in a timely manner.”
Better perception paves the way for the humanoid to climb rugged stairs and slopes continuously on the Great Wall in an unassisted walking mode.
Unlike some legged robots, which are designed to ascend and descend stairs without the aid of perception algorithms, Robot Era ensures that its robots can detect real-time changes in its immediate surroundings and identify obstacles in its tracks.
As a result, the robot is able to bypass these obstacles on the basis of real-time path planning and motion control.
“Compared to ‘blind walking,’ our robot displays better traversability and stability on complex surfaces,” Xi says.
He adds that the toughest part of the walking test stems from the design of end-to-end RL algorithms that underpin the transition from perceptive data input to robotic locomotion.
Notably, the Great Wall video came only two weeks after Robot Era unveiled XHand, a 12-degree-of-freedom dexterous hand meant for humanoids to perform various object grasping tasks.
This startup has since doubled down on investment into perceptive algorithms, reinforcement learning and neural network.
The purpose is to iterate its product portfolios at a faster speed, as well as accelerate the transfer of robot learning algorithms and models from simulation to a physical world, enabling the robot to become more versatile in real-life scenarios.
