With increasing automation in all aspects of society, humans are being tasked with interacting and collaborating with autonomous systems in a variety of contexts. At the core of successful cooperation between human and machine is trust. To that end, we are interested in mathematically modeling how human trust, and other human behaviors, evolve during interactions with automated systems and how feedback control principles can be applied to achieve the promise of automation---improving human quality of life. Through collaborations with the REID Lab here at Purdue, we have published several articles on this topic that consider novel ways of estimating human trust via psychophysiological sensing (such as galvanic skin response (GSR) and electroencephalography (EEG)) as well as using behavioral data. More recently we extended our modeling to include human workload, and we have demonstrated, experimentally, the use of adaptive user interface (UI) transparency for improving human performance when assisted by an automated decision aid. We have several ongoing projects in our group that build on our foundational work in this area. This includes human cognitive state-based control for automation designed to assist humans with skill building and modeling of human trust across multiple timescales. We are also studying how human operators make decisions in process manufacturing and using machine learning techniques to assess how automation, combined with operator expertise, can achieve better, and more consistent performance.

Our interests in controlling energy-intensive systems has led to research that addresses, in part, new algorithms and tools for controlling time-delayed systems. Twin-roll steel strip casting produces steel strips by pouring steel directly onto rollers and compressing it to a thickness near the final gauge, whereas traditional casting uses a mold to form a steel slab that is later rolled to the desired thickness. The twin roller method is 9 times more energy efficient and 7000 times faster(!) than thick slab casting. However, achieving precise physical properties along the length of the strip poses a challenging engineering problem due to the highly coupled nature of the thermal and mechanical dynamics. We developed a new algorithm, based on iterative learning control, which can compensate for time delays that are longer than a single iteration of the casting process. We were recently awarded a patent with an industry partner for our algorithm, and continue to tackle exciting controls challenges in this advanced manufacturing process.

More recently we have started investigating the role of advanced control and autonomy in future transportation systems. Through collaborations with other faculty here at Purdue and industry partners, we are designing new algorithms for platooning of heavy duty Class 8 trucks. We are also tackling large-scale modeling of the freight transportation network by considering it as a system of systems, to answer questions like “what incentives and policies, deployed over time in what way, will lead to a significant increase in the adoption of battery electric trucks?”

Research Poster(s): ►Powering What’s Next in Freight Transportation (IAC 2019)