New Robot Learns Through Trial and Error


Researchers from Berkeley, California have created a robot named “Brett” (Berkeley Robot for the Elimination of Tedious Tasks) that can perform a number of tasks and gets better at them the more times they are tried.

In order to accomplish this, his programmers reward Brett with a numerical score after he successfully completes a task.  The better he performs, the higher the number is.  “We’ve had it learn on its own, how to put caps onto bottles,” Sergey Levine said.

Rather than programming the robot for each individual task, algorithms are used that allow the robot to use trial and error to learn the tasks in much the same way as a human learns.

So far, Brett has learned a variety of tasks, including the assembly of a toy airplane, placing the claw of a toy hammer under a nail, and discovering where a square peg belongs.

While Brett currently operates in a lab setting, he is meant to one day help people out around their homes.  The challenge with that, says Trevor Darrell, co-researcher and director of the Berkeley Vision and Learning Center, is that while objects in a lab setting are always in the same position, these robots must adapt to the constantly changing environments of homes or offices.

A number of other robots like Brett have been trained to do everyday tasks including retrieving a beer from the fridge, purchase a sandwich outside the home, or even play a game of pool.

“If we can have practical household robots, we can have them go into your home, they can clean up your house, they can do the dishes, do the laundry,” Levine said.

Researchers say Brett has the capability to master tasks within 10 minutes of beginning to learn.  However, if the robot needs to find where objects are located, and thus learn extra coordinates, it could take him up to 3 hours to learn a task.

Each robot is a Personal Robot 2 (PR2) built by Willow Garage from Silicon Valley, reports Jonathan Bloom for ABC 7.

“What we’re reporting on here is a new approach to empowering a robot to learn,” said Pieter Abbeel, a professor in the university’s department of Electrical Engineering & Computer Sciences, in a prepared statement. “The key is that when a robot is faced with something new, we won’t have to reprogram it. The exact same software, which encodes how the robot can learn, was used to allow the robot to learn all the different tasks we gave it.”