Professor giving lecture

Georgia Tech Best Place to Be With Robotics Boom on Horizon, Says New Faculty Member

Animesh Garg sees a boom on the horizon for advancements in robot manipulation.

It’s an exciting time to be in the field of robotics, he said, and that’s why the new assistant professor in the School of Interactive Computing wants to be at Georgia Tech.

“The robotics faculty at Georgia Tech is particularly strong,” he said. “The reorganization and refocus of the Institute for Robotics and Intelligent Machines and the Machine Learning Ph.D. program is something that is also very special among our peer group universities.”

Garg described his research as unlocking “common sense” for robotics. In essence, he thinks carefully about mistakes that robots might make in the real world and how to preempt them before deployment.

He said building common sense into robots could be the difference over the next few years if robotics researchers are to reach a new pinnacle of achievements.

“Common sense reasoning in real-world robotics is the challenge of the next decade,” he said.
“Robots have a lot of requirements for common sense. Simple things like don’t pack glassware under heavy stuff. We should think about these problems holistically and not trying to build robots in isolation.”

Garg earned his master’s degree in industrial engineering from Georgia Tech in 2011. He also has a master’s in computer science and a doctorate in machine learning and robotics from the University of California-Berkley.

Garg spent the past four years teaching robotics, reinforcement learning, robot manipulation, and computer vision at the University of Toronto. He is also a senior research scientist at Nvidia, working on machine learning for robot manipulation.

Animesh Garg
Animesh Garg is a new assistant professor at the School of Interactive Computing who works in robotics and machine learning. Photos by Terence Rushin/College of Computing.

What interested you about coming to Georgia Tech?

The primary draw was the research culture and the school’s strength, particularly in robotics and machine learning. What is also attractive about Georgia Tech is that it’s not just computer science and engineering that are strong. There is no shortage of collaborators outside of the computer science neighborhood if I want to pursue projects in climate sciences, material sciences, or statistics.

What will your research consist of?

My research will consist of three big-picture topics. First, foundation models for representation and reasoning. How should we talk about common sense and problem solving? The second is generative AI in the context of robots so we can know more about the world through the predictive sense, which allows for better planning. The third pillar would be reinforcement — the robot learning to do something within its own abilities.

What inspired you to pursue this field of research?

Getting robots to do stuff on command is one of the longstanding science fiction challenges, right? We have used science fiction examples for many decades, but the progress has been slow. The confluence of machine learning and the ability to reason gives us the tools to solve this problem.

In the next decade, we will see more progress in robot manipulation than in the last 40 to 50 years combined. A fundamental set of problems has been solved in the last five to seven years. In the next 10 to 12 years, we will see a boom in the percolation of this technology in everyday life.

What do you hope to accomplish here at Georgia Tech?

I want to train students to be leaders in this space for the next decade, whether they choose to be in academia or start new companies. The other thing I want to establish in the research ecosystem is a consortium of robotics researchers within Georgia Tech to work closely with industry. This will enable tech transfer so what we create at Georgia Tech can be brought to fruition in industry and make a broader impact.

What are you looking forward to about teaching your students and how do you plan to work with them?

I’ll be focusing on a Ph.D. course in reinforcement learning. There is no deep reinforcement learning course offered regularly to in-person students, so that is a need I hope to fill.

The other thing I’ve been developing is a course on robot learning. The idea is that this will be a hands-on course that enables people with little background in robotics to get up to speed of a professional robotics engineer within six months.