Asst. Prof. Reza Ahmadzadeh Looks to Advance Robot Learning by Developing New Algorithms
03/13/2023
By Brooke Coupal
Asst. Prof. Reza Ahmadzadeh of the Miner School of Computer and Information Sciences envisions a world where robots can help people live more comfortably and safely, and he’s working to develop the methods to make that happen.
“In my career, I hope to get to a place where we use robots in our everyday lives,” he says.
The National Science Foundation (NSF) recognized Ahmadzadeh’s potential and awarded him a prestigious faculty early-career development CAREER grant. The nearly $500,000 grant will fund Ahmadzadeh’s project on robot learning of complex tasks over the next five years.
“This NSF CAREER award speaks volumes about Asst. Prof. Reza Ahmadzadeh’s ability to generate great ideas in the interdisciplinary field of robotics,” Kennedy College of Sciences Dean Noureddine Melikechi says.
Since his time as a Ph.D. student in robotics, cognition and interaction technologies at the University of Genoa in Italy, Ahmadzadeh has focused his research on robots learning from human demonstration. This can be accomplished through kinesthetic teaching, where a person holds the robot and moves it around, or through the robot “watching” the person via feeds from a virtual reality headset, joystick or camera.
The information gathered from the human demonstration is then sent to algorithms programmed within the robot. Existing algorithms work well for getting robots to replicate simple tasks, such as picking up an object, “but human life is not just made of simple tasks,” Ahmadzadeh says.
For his NSF CAREER project, he will be developing new algorithms for robots to learn complex tasks.
The prospective breakthroughs from this project could help automate difficult and dangerous duties in the workplace, freeing up employees’ time to pursue more creative or higher-value projects. It could also help older adults remain in their homes longer by assisting them with daily chores.
“The results of Reza’s funded research will have a great impact on how people work with robot systems in the future,” says Holly Yanco, chair of the Miner School.
For instance, the algorithms developed could give robots the ability to load a dishwasher or change a light bulb.
“If a robot is sent into a human’s home now, they will not be useful with complex tasks,” Ahmadzadeh says. “The robot could learn to hand a person a light bulb, but it wouldn’t be able to change it.”
Ahmadzadeh plans to review studies in human movement to see if algorithms used for primitive skills in robots can be improved. He then will seek new approaches that allow robots to string together a library of reusable skills to accomplish complex tasks.
Ahmadzadeh will also build algorithms that let a robot discover the new skills needed to finish a job.
“If a human shows a robot how to make coffee, the robot may know how to pick up a mug and place it, but not how to press the button on the machine,” he says. “The robot would be able to grab that skill and put it in its library to be used again later.”
Ahmadzadeh will create methods to refine the robot’s skills so it can complete a task no matter the environment. For example, if a robot is capable of opening a door, but it comes across a door with a different handle, it needs to be able to hone its skills to complete the same task.
“A human would give another demonstration, and the robot would realize that this is the same skill but applied to a new situation, so it would refine what it already knows,” he says.
Undergraduate and graduate students will be assisting Ahmadzadeh with the research, which will lead to a revamping of his Robot Learning course. Students will also get exposure to robot learning during one-day workshops that Ahmadzadeh plans to hold throughout the duration of the project. The workshops will run as part of SoarCS, a summer program for incoming first-year computer science students.
“We will teach the students the basics of robot learning algorithms, then present them with code where they can implement simple functions,” he says. “With me and the graduate students supervising, they will run the algorithm on the actual robots.”