Run by Google’s parent company, Alphabet, the X Lab has revealed details about its experimental project, named “Everyday Robot,” which focuses on teaching robots how to perform useful everyday tasks. The Alphabet robots can use cameras and sophisticated machine learning algorithms to see and learn from the world around them without the need for coding for each individual movement.
The robot uses data from its sensors to create an understanding of what it is seeing, hearing, and where it is in the world – allowing it to safely perform useful tasks among people in everyday environments.
The team is testing robots that can help in production environments. Currently, these early prototypes of robots are focusing on learning how to sort garbage. The task of grabbing something is very simple for humans, but teaching this to a robot is very difficult. Everyday Robot practice this task in both physical and virtual worlds. The laboratory has a kind of playpen, where almost 30 robots spend their daylight hours sorting garbage into trays for compost, burial, and recycling.
These operations are carried out under the supervision of human workers. At night, Everyday Robot has virtual robots practice grabbing things in simulated buildings. This simulated data is then combined with training data in the real world, which is given to the robots in a system update every week or two.
As noted in the Alphabet X’s blog post, robots are mistaken in less than 5% of cases, while company employees mistakenly throw garbage into the wrong container in 20% of cases. The engineers hope the robots can help with other, more complex tasks, like assisting elderly people in their homes.
Despite such an impressive efficiency, robots are not yet ready to replace human-sorters at garbage recycling plants. These robots are still under development. During a demonstration, one of them tried to take the air, writes The Wired.