Robot Era, a Chinese developer of humanoid robots, has tested its next-generation humanoid robot, the Star 1, in the challenging terrain of northwestern China’s Gobi Desert.
This exercise aimed to highlight the robot’s adaptability to real-world natural environments.
Two units of the flagship Star 1 were put through their paces during a running test in Yumen, a prefecture-level city in Gansu, renowned for its stunning red-sandstone Danxia landforms.
Thanks to advanced motion control algorithms, the robot exhibits impressive maneuverability across a variety of terrains, including rugged mountain trails, grassy fields, deserts, gravel, and paved roads.
During the 10-minute run, the Star 1 demonstrated its balance by slightly swinging its arms while its legs lifted off the ground, mimicking the movements of a human athlete.
The robot autonomously adjusted its speed during turns and sprints, maintaining exceptional stability throughout the run without falling or slipping, achieving a speed of 3.6 meters per second – a new benchmark for outdoor robotic runs.
Outdoor demonstrations of humanoid robots require them to sustain high dynamic balance on rough, uneven, and unpredictable surfaces, such as slopes, rocks, and gravel.
This presents a significant challenge for humanoid robots, particularly under traditional mechanical control methods. A cutting-edge solution integrates robust hardware with advanced general intelligence.
Controlling a robot’s limbs and dynamic center of gravity is essential for enabling autonomous movement on rugged terrain.
For instance, navigating soft or uneven surfaces, like those found in Danxia landforms, demands flexible, adaptive joints that can absorb impacts and adjust to surface irregularities.
Conversely, traversing hard road surfaces requires tighter joint control for stability, necessitating the robot to modulate the stiffness or flexibility of its joints for smooth movement and fall prevention.
Robot Era’s Star 1 features substantial enhancements in hardware stability compared to its predecessor. Its lower limbs are equipped with high-performance, flexible dual legs and feet, which greatly support algorithm development.
Additionally, the Star 1’s joint torque has increased from 150 N·m to an impressive 400 N·m, with a maximum joint speed of 25 rad/s, placing it among the world’s top-performing joint modules.
These advancements ensure remarkable stability and durability during high-speed movements.
Moreover, the Star 1 integrates AI and large language model technologies. Trained on an end-to-end neural network, the Star 1 quickly learns new skills and adapts to various tasks.
This versatility enables it to switch between three modes of gait – running, walking, and jumping – in the face of different terrains, including roads, grass, deserts, and uneven surfaces.
Thanks to its built-in AI model, the Star 1 supports both imitation learning and reinforcement learning. Robot Era equipped this model with extensive walking and running experience through rigorous simulation training in virtual environments.
By incorporating real human running trajectories – particularly in gait patterns and foot movements – the team achieved highly realistic, human-like motion.
Utilizing its proprietary Sim2Real technology, these lifelike movements enhance the robot’s efficiency while addressing the challenge of lower-limb dynamic stability.
Specifically, the Star 1 can run for extended periods, even when wearing shoes, without falling – a common obstacle for robots.
With advancements in algorithms and computing power, Robot Era also developed the world’s first denoising world model.
This innovation allows the robot to predict and extract key environmental information from simulation training, minimizing real-world interference with its tasks.
Consequently, the Star 1 achieves superior full-body motion control and functions effectively across diverse terrains and environments.
Additionally, Robot Era recently secured nearly 300 million yuan (approximately $42 million) in pre-Series A funding to bolster its research and development as well as commercialization efforts.
The video is below and can be accessed via this YouTube link.
