Research & Publications

Description: Human teaming with machines requires systems that can understand and adapt to their human teammates. This project incorporates objective metrics to determine the performance patterns of the human teammates (i.e., workload) in real-time, predict near term human performance and use this information to adapt the machines' interactions with the human teammates or (re-)allocate tasks among the team members in order to improve the individual teammate's or overall team's performance when completing tasks.

The work relies on using objective metrics (i.e., heart rate variability, speech rate, noise level) to assess performance. The current state of the art algorithm requires known tasks, but future efforts will focus on incorporating real-time task identification.

Sponsors: DARPA/NASA, AFOSR

I develop algorithms to extract the underlying preferences from expert demonstrations. I then use these inferred preferences to make multi-robot task allocation more effective.

Workshop Paper

Sponsors: ARL

I am improving the existing decentralized MRTA method by adding a feature to incorporate the feedback on agent trainability from the coalition level.

I am helping set up a StarCraft II user study to collect demonstrations for learning a gaussian process that will be used to improve the performance of task allocation.

Sponsors: ARL

I am extending existing algorithms to include a component that learns a personalized embedding for distinguishing heterogeneity from person-to-person differences for pose prediction.