It has been once again proved that technology has so much to learn from nature. Or at least from the way we humans communicate with animals.
The widespread method used to train dogs has been applied to robots to teach them a new multistep task at the John Hopkins University: rewarding with treats.
Lead author of the research Andrew Hundt, who is said to be inspired by rewarding his own dog Leah, thought "Why not rewarding robots to make them do something?" A dog could get anything motivating as a reward, mostly their specific ideal foods. In the case of this new robot spot, it was numeric points to be earned.
"The robot wants the higher score," Hundt explained. "It quickly learns the right behavior to get the best reward. In fact, it used to take a month of practice for the robot to achieve 100% accuracy. We were able to do it in two days."
Consider robots as newborn babies that need to be taught everything from scratch. That's what the computer scientists wanted to with Spot. Its first task was to stack blocks and Spot's actions were kind of approved or declined with numeric points. For example, his incorrect actions such as reaching out but not grasping the block earned him zero points. If it knocked over a stack, the same thing applied. In the end, Spot earned the most points when it placed the last block on top of a four-block stack.
The method called positive reinforcement helped Spot teach itself how to stack those blocks along with playing a simulated navigation game.
The training took less than what it usually takes in such cases. That is to compare a two-days time to weeks.
"At the start, the robot has no idea what it's doing but it will get better and better with each practice. It never gives up and keeps trying to stack and is able to finish the task 100% of the time," Hundt said.
The team's aim, in the long run, is to apply the findings to training household robots to wash the dishes and do laundry, which could assist the elderly with their daily tasks.
"Our goal is to eventually develop robots that can do complex tasks in the real world—like product assembly, caring for the elderly, and surgery," Gregory D. Hager, a professor of computer science, said. "We don't currently know how to program tasks like that—the world is too complex. But work like this shows us that there is promise to the idea that robots can learn how to accomplish such real-world tasks in a safe and efficient way."
Well, "Who is a good boy?" has never sounded this robotic.