In order to complete real-world tasks in home environments, offices and public places, robots should be able to effectively grab and manipulate various objects.
In recent years, developers have created various machine learning-based models designed to enable robots to skillfully manipulate objects.
, Although some of these models achieve good results, in order to perform well, they often require pre-training on large amounts of data.
The datasets used to train these models consist mainly of visual data, such as annotated images and video clips taken using cameras, but some methods also analyze other sensory inputs, such as tactile information.
According to foreign media reports, researchers at Carnegie Mellon University and Olin College of Engineering have recently explored the possibility of using contact microphones instead of traditional tactile sensors, allowing the ability to use audio data to train machine learning models for robot manipulation may open up new opportunities for large-scale multi-sensory pre-training of these models.
Related papers are published on the arXiv preprint server.
, Photo source: Carnegie Mellon University, return to First Electric Network Home>,.
