Colloquium: Interactive Intelligent Machines: Models and Algorithms for Human-Machine Interaction

Abstract: Intelligent machines are poised to transform the human experience. On the one hand, these machines (such as robots, autonomous vehicles, and digital assistants) hold an enormous potential to enhance human capabilities. On the other hand, without purposeful emphasis on their interaction with humans, they can lead to adverse side effects – ranging from operational inefficiencies to fatal accidents. The goal of my research is to realize the potential of intelligent machines by addressing foundational challenges of human-machine interaction.

In this talk, I will ground these challenges in the context of human-robot collaboration. Traditionally, robots have excelled in structured settings, such as safety cages in factories. In my research, I have designed robots that can operate beyond these structured settings and collaborate with humans in dynamic environments. I will present the models and algorithms that are central to designing these collaborative robots. First, I will discuss the need for modeling human behavior and present an approach, termed Constrained Variational Inference, to learn behavioral models with limited data and high-level manual input. Next, I will describe planning algorithms that allow robots to make decisions at execution-time and with partial information. In collaboration with industry partners, I have deployed robots to assist humans in the final assembly of cars. Based on this experience, I will illustrate that the models and algorithms for interaction not only maintain safety but also improve the efficiency of human-robot collaboration.

Finally, I will discuss future directions – which include designing human-in-the-loop learning algorithms and enabling transparency in machine’s decision-making – to realize the potential of interactive intelligent machines across domains.

Bio: Vaibhav Unhelkar is a Ph.D. candidate at Massachusetts Institute of Technology, conducting research at the intersection of artificial intelligence, robotics, and human-machine interaction. In his research, he develops models and algorithms that enable intelligent machines to enhance human capabilities. Vaibhav has published research in highly selective AI and robotics venues and, in collaboration with industry partners, deployed collaborative robots among humans. Vaibhav is a member of MIT's Work of the Future task force and has served on the program committees for AAAI symposium on Human-Agent Groups and AAMAS track on Socially Interactive Agents. Vaibhav completed his undergraduate studies at Indian Institute of Technology Bombay, where he worked on satellites and aerial vehicles.