Grab a smartphone and connect to a browser. Then sit down and teach a robot to learn basic tasks. This concept may seem too simple and just out of reach, but researchers at Stanford University are working to make teaching robots quick and easy.
The team’s approach was to create two frameworks from which robots could learn from humans instead of data or environmental exploration, which could greatly speed up the learning process. Using a robot dubbed Bender, which has a robotic arm fitted with a screen that has cartoon eyes, the researchers began teaching it with their frameworks: RoboTurk and SURREAL.
“With RoboTurk and SURREAL, we can push the boundary of what robots can do by combining lots of data collected by humans and coupling that with large-scale reinforcement learning,” said Ajay Mandlekar, doctoral student in electrical engineering and one of the framework developers.
RoboTurk makes it possible for people to direct the robotic arms in real time. It also provides background information to the robot, which helps give it a jumpstart on learning. During testing, the team used an app and browser to show Bender how to pick up steak-shaped wood.
SURREAL is designed to speed up the learning process. It runs thousands of simulated experiences simultaneously through the robot’s algorithms, helping it quickly learn from different experiences at one time.
“The twin frameworks combined can provide a mechanism for artificial intelligence–assisted human performance of tasks where we can bring humans away from dangerous environments while still retaining a similar level of proficiency in task execution,” said postdoctoral fellow Animesh Garg, a research team member.
Although not all of Bender’s lessons were successful, they do provide the team invaluable information as they continue their research. With further research, the researchers, who presented RoboTurk and SURREAL at the 2018 Conference on Robot Learning (CoRL) in Zurich, Switzerland, believe that robots could be a staple in everyday human life, assisting with chores, completing manufacturing assemblies and performing tasks too dangerous for humans.
“You shouldn’t have to tell the robot to twist its arm 20 degrees and inch forward 10 centimeters,” said Yuke Zhu, computer science doctoral student and team member. “You want to be able to tell the robot to go to the kitchen and get an apple.”
Interested in more ways machine learning is changing technology? Check out Computers Use Machine Learning to Detect Radiation Damage Better Than Humans Do and Computer Program Uses Salad-Making Videos to Learn to Predict the Future.