Recent efforts in robotics have shown robots with impressive abilities beyond what we’re used to seeing in industrial robotics. We’ve seen robots dance, climb stairs, cook Chinese food, and do gymnastics. But there are some things that robots are famously bad at, including opening doors.
Not any more. Researchers at Carnegie Mellon University have come up with a robot that can figure out how to open doors it has never seen before. It took the robot 30 to 60 minutes to figure out each one. It started out at about a 50%success rate and worked its way up to 95% by the end of the efforts.
The robot trained with a total of 20 doors, and the entire study ran about $25,000.
For the future?
We would never consider paying a human being $25,000 to spend up to 20 hours opening 20 doors. This isn’t much of an accomplishment in and of itself.
But it does hold out hope for future developments. Robots that can figure out how to solve problems would be a major step up. Being able to figure things out without human intervention would make most robots more useful. And opening doors has for many years been a major obstacle to using robots in search and rescue situation, among other tasks.
Why is it so hard for robots to open doors?
Hardware limitations are one of the issues. Many robots, especially industrial robots, have limited dexterity in their manipulators. This makes it difficult for them to grasp and manipulate the doorknobs, handles, or levers that open doors.
Some robots may not have the strength necessary to turn heavy door handles or push against heavy doors. Robots rely on sensors to perceive their surroundings. If their sensors are not advanced enough to accurately perceive the location and orientation of the door and its opening mechanism, they may struggle to interact with it successfully.
Software limitations are the other half of the equation. Robots need to be programmed with specific instructions on how to open doors. Since there are so many different types of doors and related mechanisms, it’s a complex task requiring understanding of physics, perception, and motor control. Being programmed to open one door doesn’t mean the robot will be able to open any other doors. That’s where the new work at Carnegie Mellon comes into play.
Industrial robots probably don’t need to open doors. We can see the value for search and rescue, service robots such as robot waiters, nurses, and office helpers, and robots designed to provide elder care or child care.
But the possibilities goes way beyond just opening doors. Robots that can figure out how to approach new challenges on their own will be able to overcome significant limitations in every use case.