A Robot that Automatically Finds and Opens a Door When Charging

A team of engineering students at the University of Cincinnati is building an autonomous robot that can open a door, find the nearest electrical outlet, and charge it without human assistance.

A Robot that Automatically Finds and Opens a Door When Charging

Door - Robot Kryptonite One of the biggest obstacles for robots is the door. Ou Ma is a Professor of Aerospace Engineering at the University of Cincinnati. “Robots can do a lot of things, but getting one of them to open the door and let it go through is a big deal,” said Ma. The team has overcome these challenges in 3D digital modeling, a huge step forward for robotic assistants. These robots could include robots that vacuum and disinfect office buildings, airports, and hospitals. They make up a significant portion of the $27 billion robotics industry.

Yufeng Sun is a senior research author and Ph.D. student at the University of California School of Engineering and Applied Sciences. Some researchers have solved the problem by scanning the entire room to create a 3D digital model that allows the robot to find the door, Sun said. However, this is a tedious solution that only applies to scanned rooms.

There are many challenges in developing autonomous robots that open doors on their own. First, it comes in a variety of colors and sizes and has a variety of handles that can be lower or higher. The robot also needs to know how much force it must use to open the door to overcome the resistance. Because many public doors close automatically, the robot may lose its grip and have to start over.

Using Machine Learning

Using machine learning, UCLA students were able to "teach" themselves how to open a door by a robot through trial and error. This means that the robot can correct mistakes on the fly and prepare for real work through simulation. “The robot needs enough data or 'experience' to train,” says Sun. “This is a big problem for other robotics applications that use AI-based approaches to solving real-world problems.”

Sun and UCLA graduate student Sam King transform successful simulation studies into real robots. "The question is how to translate this learned governance policy from simulation to reality, which is often referred to as the Sim2Real problem," Sun said.

Another problem is that digital simulations are typically only 60-70% successful in initial real-world applications. Therefore, Sun plans to spend at least a year perfecting the new autonomous robotic system.