Feedback Control and Human-Agent Teaming


Monday, November 20, 2023 - 3:30pm to 4:30pm


ESB 1001


Professor Vaibhav Srivastava

In this talk, I will delve into various challenges related to human-agent teaming and analyze them through the lens of control-theoretic techniques. I will begin by examining human motor learning in complex, high-dimensional tasks—a fundamental issue in the post-stroke rehabilitation of high-degree-of-freedom systems like the human hand. To address this, I will introduce a motor learning model grounded in the concept of "synergies" derived from the motor learning literature, coupled with adaptive control techniques. This model not only explains experimental data but also quantifies several learning trade-offs that were previously difficult to quantify. Moreover, I will demonstrate that the parameter fits for human data lie at the precipice of a phase transition concerning these trade-offs. I will also outline ongoing initiatives aimed at leveraging this model to expedite motor learning. Subsequently, I will shift focus to our endeavors in trust-aware human-robot teaming. Specifically, I will explore a scenario where a robot, operating in a semi-autonomous object collection setting, can seek human assistance. Employing a POMDP-based approach that treats human trust as a hidden variable, and seeks to optimize team performance. Results from human experiments will be presented, showcasing the superiority of a trust-aware policy over a trust-agnostic one in terms of overall team performance. If time permits, I will highlight some of our other efforts in the domain of human-in-the-loop systems, including, understanding the impact of evaluative feedback on the learning and skill transfer performance of a human engaged in sequential decision-making tasks.

Event Type: