Pssst! How to make Sony's robot dog Aibo run more quietly
Robots used in the home should be as quiet as possible. This also includes quiet running.
(Image: ETH ZĂĽrich (Screenshot))
Researchers at the Swiss Federal Institute of Technology in Zurich (ETH Zurich) and Sony have used reinforcement learning (RL) to teach Sony's robot dog Aibo to walk more quietly. Many users of the robot dog complain about the loud noise it makes when walking.
In principle, loud noises are a general problem with robots. Even though the majority of them now rely on electric actuators, they still make noises when they put their feet on the floor, for example. Robots that are used in the home in particular should do their job as quietly as possible without disturbing the occupants.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
Optimization of walking noises
The scientists at ETH Zurich and Sony rely on reinforcement learning for their noise reduction approach. In a physics simulation, they tried out various movements of the joints of the robot dog Aibo. Movements that promised reduced noise were rewarded during virtual training, while other movements were penalized. The main aim was to minimize the speed of foot contact to reduce the noise when the foot touched down, the researchers write in the study "Learning Quiet Walking for a Small Home Robot", which is published in the preprint on Arxiv.
Using the training, the researchers were able to control each individual joint in such a way that it is damped and stiffened. They tested the approach with Aibo in the real world. They used additional sensors in the feet of the real Aibo. The data collected was used to penalize noises caused by rapid movements, for example. A microphone at the back of Aibo's head recorded the respective noise development.
Videos by heise
In real experiments with Aibo, the researchers determined the noise generated when walking with the original movement algorithms and the noise generated when walking with the algorithms optimized by the scientists. The comparison showed that RL-based running produced significantly less noise. However, there was also a decline in performance when walking. For example, the robot was no longer able to climb ramps with steep inclines. The researchers therefore adapted the walking movements in such a way that they resulted in a compromise between loudness and robustness.
The scientists assume that their results can also be transferred to other robots used in the home.
(olb)
