Human-in-the-Loop Automation

We are developing human-in-the-loop interfaces for task-based assistance and training. Using these interfaces, we address issues of uncertainty from both the human and unknown dynamics of the environment while allowing users to be as autonomous as possible.  This work moves away from controls that prioritize trajectory error that use a priori knowledge of the joint human-machine system in favor of data-driven approaches with applications to a broad class of tasks and sensorimotor deficits.

Controllers as Filters: Human-in-the-loop Interfaces base on Maxwell's Demon

This research focuses on the development of control algorithms that can handle input signals with high uncertainty and noise. Using an optimal control as a point of comparison, we can intelligently filter Gaussian noise or human inputs to improve success or performance of a given task. The concept is based on Maxwell's Demon and can be used to synthesize human-in-the-loop interfaces that enhance task learning.

Watch the NSF "Science Nation" feature video of this work here.

Optimal Tactile Feedback to Enhance Learning and Rehabilitation

This project uses optimal control to provide synthetic sensory feedback to promote learning or re-learning of sensory motor skills. Specifically, we use vibrotactile feedback to encode an optimized linear combinations of state information, providing information about the optimal course of action. We have studied this in healthy subjects performing a balancing task as well as impaired subjects performing a tracking task. In both cases, task performance was enhanced by a combination of visual and tactile input.

Related publications