Create your CubeMars account
In complex terrain exploration and extreme rescue scenarios, the vertical climbing capability of quadruped robots has always been a major technical challenge.
In this video, we showcase the highly innovative KLEIYN quadruped wall-climbing robot. By introducing an Active Waist Joint and deeply integrating Reinforcement Learning (RL) algorithms, this project successfully achieves efficient vertical chimney climbing inside narrow walls with widths of 800–1000 mm, reaching a climbing speed 50 times faster than traditional quadruped robots.
In this breakthrough research, CubeMars high-performance joint actuators serve as the core power solution, providing essential high-dynamic response and precise force-control support for KLEIYN’s waist and leg joints.
Traditional quadruped robots usually only have degrees of freedom in their legs, while their torso remains rigid.
This design provides excellent stability on flat terrain, but when facing vertical walls or narrow spaces such as chimneys and pipelines, it encounters significant physical limitations:
Lack of a torso-based force generation point: A rigid torso cannot effectively transfer leg forces to the wall through movements such as “arching the back” or twisting, resulting in insufficient friction and support force.
Limitations of control algorithms: In confined spaces, robots require extremely frequent posture adjustments. Traditional control methods struggle to handle such high-dimensional and nonlinear dynamic problems.
To overcome these challenges, the KLEIYN team innovatively added an active waist joint to the robot and adopted a reinforcement learning-based control strategy.
CubeMars joint motors provide ideal support for this project in the following key aspects:
Providing powerful and agile drive performance for the “active waist”: The active waist is the key component that enables KLEIYN’s wall-climbing capability. It needs to frequently perform large-angle pitching and yawing movements. CubeMars joint motors feature high torque density and extremely low rotor inertia, enabling them to deliver powerful explosive force instantly while maintaining a high response bandwidth.This allows the robot’s waist movements to become as flexible and dynamic as biological motion.
Perfectly supporting the low-level execution of reinforcement learning (RL) algorithms: Reinforcement learning algorithms require actuators with extremely high control accuracy and minimal latency. CubeMars motors integrate high-resolution encoders and advanced FOC control algorithms, enabling precise execution of position, velocity, and torque commands from the upper controller at high frequencies (such as 1000 Hz). This ensures accurate reproduction of RL policies in real-world environments.
Extremely lightweight design:For wall-climbing robots, every additional gram of weight significantly increases the requirements for motor torque and adhesion/support mechanisms. CubeMars motors integrate the driver, motor, and gearbox into a highly compact unit with lightweight construction.This helps KLEIYN add additional waist degrees of freedom while strictly controlling the overall robot weight.
Through this video, you can directly observe the impressive performance of the KLEIYN robot powered by CubeMars motors:
Narrow wall climbing test: Demonstrates how the robot uses its active waist to “brace” against walls and climb steadily upward at a speed of 150 mm/s inside vertical walls with widths of 800–1000 mm.
Active waist motion close-up: Clearly shows the flexion, extension, and twisting movements of the active waist during each climbing step — motions that traditional rigid quadruped robots cannot achieve.
Comparison with traditional solutions: Visually demonstrates KLEIYN’s overwhelming advantage over traditional quadruped robots in climbing efficiency.
Q1: Are CubeMars motors suitable for robot waist joints?
A:Yes. CubeMars AK Series joint modules feature high torque, low backlash, and high-frequency response characteristics.They are widely used in waist, neck, and joint connection structures of humanoid robots and quadruped robots, providing strong support for complex whole-body dynamic control.
Q2: Do your motors support advanced control algorithms such as reinforcement learning (RL)?
A:Yes. CubeMars motors provide high-frequency CAN/RS485 communication interfaces, support multiple control modes including position, velocity, and torque control, and offer open low-level communication protocols. They can seamlessly integrate with various reinforcement learning frameworks such as Isaac Gym and MuJoCo, enabling rapid Sim2Real (simulation-to-reality) deployment.
Whether you are developing a quadruped wall-climbing robot, humanoid robot, or complex robotic arm, CubeMars provides miniature actuator solutions with high power density and fast response speed tailored to your needs.
[Learn more about CubeMars AK Series joint actuator specifications and dimensions]
