AI to the Rescue

AI to the Rescue

How Robots Learn in Dangerous Environments

A new research project that teaches robots to rapidly adapt in unknown situations could help search and rescue workers in dangerous environments, while saving more lives in the process according to Xiangnan Zhong, Ph.D., an assistant professor in the College of Engineering and Computer Science.

Her work in this field recently earned her a $503,000 CAREER grant from the National Science Foundation (NSF), which is NSF's most prestigious award in support of early-career faculty.

"This project combines research and education, two key components that will prepare future generations in the fields of AI, learning and control," Zhong said.

With this award, Zhong said her goal is to initially enable artificial intelligent (AI) robots with basic skills and then let them automatically learn comprehensive skills based on the information they have mastered and eventually solve complex problems.

For example, one of her tests requires AI robots to find specific targets in a complicated and highly dynamic maze with lots of uncertainties that they never met before. This kind of task is hard to be directly learned by the AI robots, she said.

Zhong breaks down the learning into multiple steps. The robots start to learn the basic skills, such as avoiding hitting objects or finding the charging outlet. Then, they will learn how to find the fastest way to the targets. Beyond this, they will be also required to deal with the dynamical environment, such as changing path if there are unexpected obstacles shown in the original planned path, Zhong said.

Zhong uses a computer software system to teach her robots. She equips them with sensors, such as cameras. Once the sensors are placed, the robots can see an image of the environment, understand the distance from the obstacles and know the temperature of their surroundings, she said.

The intelligent learning method continues until the robots can successfully adapt to random environments without needing assistance. This learning process is based on the reinforcement learning strategies where the robots are rewarded when behaving correctly, for interacting physical systems when the robots with which they interact may react in inconsistent ways, Zhong said.

For Zhong, this new robot learning experience is similar to how humans learn and adapt. "When we grow up, we can learn from the interaction with our environment. We learn the basic skills first, and then the complicated skills are built up based on the basic ones," she said.

She compares the maze task to a person getting lost inside the mall without seeing the exit in sight. The same context clues that people use to find their way out or to remember where a car is park, are the same strategies applied to her robots, she said.

"If an AI robot can learn and become autonomous and intelligent like humans, they can help rescue workers in dangerous situations where humans cannot enter, such as a searching for a lost family in a melting ice cap or rescuing someone from a collapsed building," Zhong said.

Zhong's research on learning methods for connected AI systems has been longstanding. Some of her previous work included autonomous vehicles design, cyber physical systems development to prevent and detect cyberattacks and communication techniques design for robots' collaborations. Her work has also been funded by NSF Computer and Information Science and Engineering Research Initiation Initiative Program and NSF Energy, Power, Control and Networks Program, she said.

When Zhong became interested in this field, she was a doctoral student at the University of Rhode Island in 2012. There she studied electrical engineering and how to design intelligent complex systems.

Once she became an assistant professor in 2017 at the University of North Texas, Zhong said she immediately dug into AI learning and intelligent systems to begin her journey of enabling robots to learn and adapt in unknown environments.

"I'm honored by this award and that it recognizes my career and the long-term goal I've had," Zhong said. "I know the work that I'm doing will be valuable to society."

If you would like more information, please contact us at dorcommunications@kome-shibahara.com .