Reliable Autonomous Systems Lab

PhD Student: Michael Lee
Undergraduate Students: Vignesh Rajmohan

Our capacity to deploy, collaborate, and co-exist fluently with robots is contingent on our ability to understand their decision-making. For example, we likely wouldn't rely on them until we could reliably predict their decision-making and subsequent behavior in new situations. Thus we are designing agents that can teach humans their decision-making through demonstrations. We model humans as inverse reinforcement learners and leverage human learning strategies (e.g. scaffolding) to select demonstrations that are both informative and easily understood by humans.

In this research, we assume that the robot’s decision-making is captured by its reward function, comprised of reward features known to a human learner and reward weights unknown to the human learner. Thus we seek to convey the unknown reward weights to a human through demonstrations and inverse reinforcement learning.

PhD Student: Pallavi Koppal

Autonomous agents need to learn increasingly competent and complex behaviors. One way of effectively learning these behaviors is to include people in the learning process. Therefore, we are investigating human-in-the-loop strategies that are both more user friendly and lead to efficient learning.

PhD Student: Roshni Kaushik
Undergraduate Students: Jaehee Kim, Peter Vanegas, Allison Moore, Mark Chen, Priyanshi Garg, Bharath Sreenivas, Adrian Thinnyun, Steven Qu

Peer tutoring, in which one student teaches material to another student, has been shown to increase learning. In particular, there can be significant learning gains for the person taking on the role of the teacher if they can engage in reflective knowledge building as they tutor. We propose that a suitable social robot could believably play the role of ignorant learner, while in reality it would already understand the concept being taught and so could subtly guide the student teacher with its behaviors and modes of interaction (often referred to as “back leading”). Just as important as helping students cognitively is to help them emotionally: providing encouragement and helping students maintain focus, build confidence, persevere, overcome frustration, etc.

PhD Student: Michelle Zhao
Undergraduate Students: Timothy Hyun

In order to develop AI partners that can become effective teammates, we must endow them with basic social understanding—particularly the ability to understand humans' intentions, knowledge, and mental states. The ability to infer other humans' mental models and use those models to perform actions, make predictions, and evaluate outcomes is called Theory of Mind (ToM). This research focuses on developing algorithms for artificial agents to be able to collaborate with human partners by (i) understanding human behavior and preferences, (ii) coordinating on task strategies, and (iii) influencing mutual adaptation between agent and human. Using integrated ML approaches, including Reinforcement Learning and Behavioral Cloning (Imitation Learning), we aim to build effective robot and AI teammates with the ability to both understand the strategic preferences of their partner, and communicate their own preferences.

Masters Student: Daphne Chen

In order for artificially intelligent agents to effectively work with humans in assistive settings, it is necessary for them to understand individual preferences for accomplishing different tasks. Thus in this work we aim to develop personalized models of human behavior that can learn how a human accomplishes a task within the constraints of their environment.