A groundbreaking advancement in robotics could revolutionize product supply chains by using fast-moving autonomous mobile robots to deliver goods efficiently and independently. Researchers at ETH Zurich’s Robotic Systems Lab have introduced a cutting-edge robot design that combines the capabilities of both wheeled and legged robots. This innovative robot, featured in a recent Science Robotics paper, utilizes various reinforcement learning techniques to seamlessly transition between driving and walking modes, adjusting to different terrains.
Joonho Lee, one of the co-authors of the paper, explained that the project’s primary goal was to create a large-scale autonomous driving system for a ground robot with unmatched speed, achieved after over five years of extensive research in legged robotics, autonomous navigation, and robot perception.
Building on the success of a previous robot developed by team CERBERUS, the ETH Zurich team’s robotic system features a simplified design and a more advanced AI-powered navigation system. By developing and training hierarchical reinforcement learning techniques, the researchers trained a neural network-based controller capable of generating navigation plans within milliseconds, allowing the robot to autonomously navigate diverse environments efficiently.
The unique advantage of this approach lies in the system’s comprehensive understanding of the complex dynamics of legged robots, enabling it to navigate the robot effectively across various terrains. This versatility allows the robot to leverage its wheels on smooth terrains for optimized power consumption and seamlessly switch to walking mode to navigate challenging terrains with obstacles.
With the ability to combine the strengths of wheeled and legged robots, the controller developed by Lee and his colleagues offers a significant improvement over traditional planning and control techniques. By leveraging advanced technologies and innovative approaches, this autonomously navigating wheeled-legged robot sets a new standard for efficient and effective robotic delivery systems.
For more information on this groundbreaking research, refer to the publication “Learning robust autonomous navigation and locomotion for wheeled-legged robots” in Science Robotics.
