Robots take baby steps to walking like humans
With the help of human behaviour theory, a new robot has been developed, which knows how to balance better than any robot so far - and maybe humans could learn from this theory too..
Applying theories of animal and human behaviour to robot technology, a psychologist working with robot scientists at Manchester University is helping a robot balance better.
“Robots are yet to match the abilities of those in science fiction hits like Star Wars and Blade Runner, and none have mastered walking on two feet,” says psychologist Warren Mansell, lead author of this study.
Indeed, robots do not balance very well. Unless they have really big feet, or no feet at all, they struggle to stay vertical. “The first step to making a robot walk is to make it balance efficiently,” explains Mansell.
The team tested identical robots, assembled from Lego parts and sensors, using two different pieces of software: the typical software used by engineers, and a new software based on a theory of behaviour called "Perceptual Control Theory". The robot’s balance was markedly better with the latter. When pushed or shoved, it returned to its upright position faster and more smoothly.
Current software, which tells a robot how to walk, plans every step ahead: it pre-calculates the robot’s orientation, angle of the wheels and aerodynamic forces at several future timepoints. The new software, in contrast, focuses on keeping the robot vertical by making it follow goals such as to move at a certain speed at a given time.
Mansell compares this method to standing one leg: “When you’re standing on one leg, your knee is twitching all over the place, as your muscles are contracting to maintain balance; but you are not concentrating on that; rather, you’re looking at one fixed point.”
This software makes the robot “sense only what it needs to sense.” The focus on the outcome allows the robot to find ways of balancing that could not have been programmed in advance.
The new method could enable robots to work autonomously in the future, as it helps them learn how to behave based on controlled outcomes and their perceptions of the environment.
The outcomes of the study not only get robots to balance better and more efficiently. They also simulate a type of reasoning, which could help humans make better decisions.
Using desired outcomes as a starting point is a universal method - Mansell uses the same theory to help his patients "identify what it is that they want, rather than how to get there, to help them control their lives better and balance priorities in different ways," he says.