Inspenet, October 30, 2023.
New robotic learning algorithm
A group of scientists at the Georgia Institute of Technology has developed a method to identify a person’s location by detecting subtle sounds that emanate naturally from the human body, such as the sound of walking.
This novel approach is powered by machine learning algorithms and is based on a data set comprising 14 hours of high-quality four-channel audio recordings, combined with images captured by a 360 RGB camera.
Importantly, these recordings were obtained through experiments in which individuals were asked to move around a robot in various shapes and scenarios. Consequently, the researchers trained this model in order to filter and discard external sounds that were not relevant, such as those coming from heating or cooling systems, as well as sounds emitted by the machine itself.
“Robots typically use cameras or LiDAR to navigate around people, but if those sensors fail or are unavailable, our method allows them to rely solely on audio, which is typically already available in most hardware configurations,” they explain. .
Initial experiments were carried out using the Stretch RE-1 and revealed that this approach demonstrated two times the performance compared to other sound-based localization systems.
“ This can be potentially useful for robots that navigate indoor spaces shared with people, allowing a non-intrusive method to detect where people are ,” they conclude.
In the future, it is intended to use this localization technique to increase safety and optimize the performance of robots intended to work closely with people.