Computer Science Asst. Prof. Reza Ahmadzadeh Collaborates with WPI, Harvard on Amazon Robotics-Funded Project

Reza with robots Image by Brooke Coupal
Asst. Prof. Reza Ahmadzadeh is developing algorithms that allow robots to learn from humans.

01/18/2023
By Brooke Coupal

Asst. Prof. Reza Ahmadzadeh of the Miner School of Computer and Information Sciences envisions a future where more people have robots inside their homes completing housework and other routine tasks.

But before that becomes a reality, Ahmadzadeh says robots need to be more user-friendly, so that people without programming skills can successfully use them.

Under a nearly $500,000, one-year grant from Amazon Robotics, Ahmadzadeh is working in collaboration with Worcester Polytechnic Institute (WPI) and Harvard University to develop algorithms that allow robots to learn from humans without the need for programming, making it easier for people to teach a robot how to do household chores, such as making a cup of coffee or folding laundry.

“The goal is to teach the robot through human demonstrations, which will eventually be very useful for non-experts that have robots at home,” Ahmadzadeh says.

Humans Transfer Skills to Robots

WPI principal investigators Berk Calli, Jane Li, Cagdas Onal and Haichong Zhang are developing a glove made of sensors that will be worn by human operators as they perform different tasks. The sensors will record the amount of force the operator is using as well as the positions of their finger joints.

Ahmadzadeh, who is the sole principal investigator for UMass Lowell, will take this information and build algorithms that allow the robot to mimic the human operator’s movements.

“We’re not coding a step-by-step movement of the robot,” he says. “We’re building a family of algorithms that model tasks as it sees them.”

The algorithms will be inputted into the robot’s computer, which will prompt the robot to complete the task autonomously. If the robot is unable to do so, a failure detection system, being created by Harvard principal investigator Robert Howe, will alert the human user. From a remote position, the human will be able to use the glove to show the robot another example of how to do the task. The information from the glove’s sensors will be sent directly to the robot through either the internet or a wired Ethernet connection, allowing the algorithms to update their model of movement in real time so the robot can finish the task.

“As we see examples of failure, we can refine the robot’s skills so it becomes better,” Ahmadzadeh says.

Reza robots Image by Brooke Coupal
Ahmadzadeh's lab houses robots that will be used in the Amazon Robotics-funded project.

Robots Offer Benefits to Humans

Ahmadzadeh sees robots as a way to help people live independently, especially as they age. About 17% of the United States population, or 56.1 million people, was 65 years or older in 2020, according to the U.S. Census Bureau. This number is projected to climb to 94.7 million people, or 23% of the U.S. population, by 2060.

“Our research project will help the elderly as they gradually lose the capability to do things,” he says. “Eventually, they can use the glove to teach the robot how to make coffee, for instance, and then they can rely on the robot once it learns that task.”

Amazon and other companies are increasingly using robots in warehouses and on assembly lines to free workers from some tasks. Employees equipped with the sensor glove would be able to teach new skills to robots outfitted with Ahmadzadeh’s algorithms. The robot would then be capable of completing additional tasks, such as picking up clutter.

Organizations are also progressively utilizing robots in places that are dangerous for humans. With the research from Ahmadzadeh and his team, a robot entering an unsafe place could learn instantaneously how to fix an issue at hand. For example, if a nuclear power plant is experiencing a leak, a robot could be sent inside with a camera to observe the problem. A person outside the plant watching the camera’s feed could then use the sensor glove to teach the robot how to repair the leak.

The researchers plan to share their findings at local scientific events, including the Botfest in Lowell, the Cambridge Science Festival in Cambridge, the MassRobotics Robot Block Party in Boston and TouchTomorrow in Worcester.

Summer workshops will be made available to undergraduates, where they will learn about the project and experiment with the robots. Ahmadzadeh will also be hiring a graduate student to assist with the research.

“This will give the students valuable hands-on experience,” he says.