Artificial Intelligence (AI) and robot servants are not that far from becoming a reality. In an attempt to eliminate tedious daily tasks from our modern lives, researchers at the University of California-Berkeley have developed a robot that has evolved the ability to learn how to perform various tasks on its own, through trial and error. It’s the exact process that the human brain uses as well.
To prove Brett’s, the newly developed robot’s, learning capacities the researchers released a video of it completing several tasks. The clip begins with the robot attaching a wheel part onto a toy airplane. Brett starts by randomly pocking the plastic airplane with the wheel. It hits the wall several times before the wheel part finally connects with its proper slot on the toy.
After it first manages to attach the wheel properly, the robot realizes on its own where the wheel part needs to be connected and repeats the process several times without any further error.
The robot is then shown assembling a Lego block and the video informs that for Brett, this is the human equivalent of inserting a key into a keyhole. It is also shown cleaning shoes, putting caps back on top of bottles, and stacking hoops on top of each other on a wooden stand.
This is a technological breakthrough in developing and applying learning algorithms. Brett did not have any pre-programmed details on the task that it would be completing, or on how to navigate them. Instead, as Pieter Abbeel, associate professor in the campus’s electrical engineering and computer sciences department, explains “What we’re reporting on here is a new approach to empowering a robot to learn”.
Trevor Darrell, co-researcher, further adds that “The challenge of putting robots into real-life settings, like homes or offices, is that those environments are constantly changing. The robot must be able to perceive and adapt to its surroundings”.
The team’s goal was to not have to reprogram the robot whenever it’s faced with doing something new. The branch that the duo focused on is called “deep learning”. What it basically means is that layers of artificial neurons process overlapping raw sensory data. This allows the machine to categorize sound waves and image pixels, and to recognize patterns and categories in the data that it is receiving.
The same software that encodes how the robot is capable of learning anything at all is also the same software that allows the robot to learn how to perform each task in particular.
Deep learning has been inspired by the neural circuitry of the human brain, which means that Brett relies on personal experience in order to acquire new skill sets throughout “its life” and to learn how to navigate a 3D environment. It does not require conventional preprogramming.
The robot can complete a new task in about ten (10) minutes if it’s given the relevant coordinates for where a task begins and where it ends. If it’s not given any coordinates whatsoever, it takes roughly three (3) hours to coordinate vision and control, and to learn the process.
Design wise, Brett si not much to look at. The robot is not very polished, stylish or futuristic-looking. If anything, it looks quite retro, like something movies used to come up with in the 80’s. In fact, the researchers used a Willow Garage Personal Robot 2 (PR2) for their experiment.
Image Source: pddnet.com