Driving or walking? Potential delivery robot chooses its own locomotion
Depending on the terrain, a robot chooses whether to drive or walk. This could give it advantages as a delivery robot.
The ETH robot can drive or walk - depending on the obstacle to be overcome.
(Image: Joonho Lee / ETH Zürich)
A research team from the Robotic System Lab at the Swiss Federal Institute of Technology in Zurich (ETH Zurich) has developed a robot with legs and wheels. The scientists have used reinforcement learning to teach it to switch automatically and seamlessly between driving and walking depending on the terrain. Such a robot could be used as a delivery robot, for example, enabling it to overcome stairs and other obstacles in comparison to its exclusively wheeled colleagues.
The wheel-legged floor robot is based on an earlier robot developed at ETH Zurich, the researchers write in their scientific paper "Learning robust autonomous navigation and locomotion for wheeled-legged robots", which was published in Science Robotics. This earlier version won the 2021 DARPA Subterranean Challenge. The researchers greatly revised the design of this robot, simplifying it and giving it a more advanced artificial intelligence (AI)-based navigation system.
"Traditionally, navigation planning for ground robots has been done using online optimization methods", explains Jonhoo Lee, one of the authors of the study. "Such approaches work fine for simple wheeled robots or slow-walking robots, but in the case of fast-moving robots like ours (which can drive up to 20 km/h) they cannot provide fast enough navigation plans. For robots moving at 2 m/s, 0.5 seconds of delay can result in a 1 m error, which can lead to a catastrophic collision."
To enable the robot to navigate autonomously at high speeds of up to 20 km/h, the scientists developed various hierarchical reinforcement learning techniques, which they tested and evaluated. They used the best approach to train a controller based on a neural network. The controller can process different types of input and generate new navigation plans for the robot within a few milliseconds.
The neural network controller can understand the complex dynamics of legged robots, says Lee. Accordingly, the robot can navigate very efficiently on different surfaces at different speeds. For example, the ETH wheel-leg robot uses its wheels on flat surfaces. This allows it to move faster by rolling – while at the same time-consuming less power. On more difficult terrain with obstacles, such as steps, it then switches to walking mode.
The neural network developed by the researchers processes data supplied by the robot's sensors. This allows the most efficient method of locomotion to be determined and the advantages of wheel-based robots to be combined with those of legged robots.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmmung wird hier ein externes Video (Kaltura Inc.) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Kaltura Inc.) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
"Wheeled robots are efficient but cannot traverse high obstacles", says Lee. "On the other hand, legged robots are very good at overcoming obstacles and steep slopes, but their efficiency is very low because they have to drive more than 10 joints in an irregular pattern. Usually, walking robots can only operate for up to 1 hour. With the wheeled legs, our robot can overcome the same obstacles as normal walking robots with at least 3 times longer operation."
Specialized neural networks
The wheel-leg robot uses a total of two neural networks. One exclusively controls the robot's movements. The second, on the other hand, concentrates exclusively on navigation. In combination, this results in a very robust overall system that can cope with rough terrain, even in labyrinthine environments. The scientists needed less than a year to develop the system.
The ETH Zurich scientists tested the robot in the two cities of Zurich and Seville by having it drive and walk through a real urban environment. The robot covered a distance of over 10 km autonomously and was able to automatically adapt to different terrain formations.
The researchers now want to further improve the system. They see the robot's most promising application as the autonomous delivery of goods. The robot can overcome almost any obstacle so that customers do not have to go to special pick-up points, as is the case with conventional delivery robots, which cannot climb stairs with their wheels and therefore cannot reach every front door.
(olb)