Robotics Program Introduces the Everyday Trash-Sorting Robot

The X Lab is teaching robots how to complete everyday tasks.
Fabienne Lang

You've most likely seen or watched the impressive advances made in the robotics field in the past few decades. That said, robots are still a long shot away from performing day-to-day helpful tasks.

This is where the X Lab enters. Run by Google's parent company, Alphabet, the X Lab is currently working on an experimental project that focuses on teaching robots how to perform useful tasks.


In a blog post, general manager of the entire project, Hans Peter Brondmo, wrote that the company's engineers would now be focusing on creating robots that can interact with humans in meaningful ways, as well as perform handy tasks.

The Everyday Robot Project

The first trick that the engineers from the X Lab have focused on is one that most humans don't wish to perform, which is sorting out trash. 

Robotics Program Introduces the Everyday Trash-Sorting Robot
The Everyday Robot in the X Lab office. Source: X Lab

Alphabet got their idea of creating a trash-sorting robot as the team noticed some of the recyclable or compostable trash placed in the wrong bins in their office. Unfortunately, a lot of this waste ends up in landfills, and can't be properly recycled. 

So the engineers took the matter into their own hands and decided to teach robots to go through trash items, moving them from the wrong bins into the right ones.

In the traditional way of teaching robots new maneuvers, you would code the robot to recognize certain items and move them. 

In X Lab's moving images below, you can see the robot improving its sorting ability — starting from the image on the left to the right. 

Robotics Program Introduces the Everyday Trash-Sorting Robot

The X Lab wanted to try something novel. Instead of applying code, they decided to use simulation, reinforcement learning, and collaborative learning. This is how they did it: 

Most Popular

Once the office was closed, virtual robots practiced sorting virtual rubbish into virtual bins in a virtual office. This was then taught to real robots doing the actual job. Then, what the daytime real robots learned would be passed on to the nighttime virtual robots, who adapted and practiced some more.

The outcome was impressive: these robots were able to learn the tasks at hand not through code, but through virtual learning and practice.

Moreover, their work was successful. The office waste contamination went from 20% down to under 5%

So what's next for the robots? The team will keep developing them, and see if they can teach the robots to use their transferable skills in other useful tasks, without having to input any code.

The hope is to create robots that can properly assist with our daily complex tasks.

message circleSHOW COMMENT (1)chevron