Meeko Mitsuko K. Oishi, PhD
Hybrid Systems and Control Lab
Reachability Analysis and Controller Synthesis
Reachability analysis is powerful in its ability to provide assurances of safety despite bounded control authority and disturbances. However, for generic systems and constraints, computing the minimal reachable set (equivalently, the viability kernel) can be prohibitively expensive. We investgate techniques to improve the computational cost for certain classes of systems. Some of the techniques we have developed are based on structure decomposition, and involve solving multiple smaller reachability calculations. Other techniques use the efficient Lagrangian methods available for minimal reachable set calculation, by iteratively computing the maximal reachable set over small time horizons. With these techniques, viable sets can be computed for systems with several tens of states.
Stochstic reachability analysis provides a minimum likelihood of safety, despite bounded control authority and stochastic disturbances. Computing stochastic reachable sets is also very computationally expensive. We have begun to investigate methods to compute stochastic viable sets for moderate dimensional systems. We are also developing optimal controllers for systems under incomplete information.
We have applied these techniques to problems in safe delivery of automated anesteshia, aircraft flight management systems, and space vehicle docking.
Observability of Human-Automation Systems
In large, complex human-automation systems, succinct user-interfaces are key for effective human-automation interaction. However, the user-interface may be misleading if insufficient or incorrect information about the underlying system is provided to the user through the user-interface. We consider the user-interface to be an output map to a hybrid dynamical system, and pose the question of adequate information in the user-interface as one of observability. The user is a special type of observer, with additional constraints that model human limitations of information processing, as well as guidelines for "good" human-automation interaction. We are interested in creating theory and computational tools for both analysis as well as design. Our current work has foucsed on visual modalities, but we also plan to consider auditory and haptic information streams, as well.
Collaborative Control of Human-Automation Systems
In semi-automated systems, mismatches between automation intent and user intent can cause "fighting" and other undesirable behavior. We are extending reachability-based tools to provide assurances of safety in human-automation systems (determinstic and stochastic).
Autoregulation in Traumatic Brain Injury
Monitoring intracranial parameters such as cerebral blood flow, and identification of prognostic cerebral blood flow dynamics may lead to improved treatment earlier in recovery and may further aid prevention of secondary injuries. In collaboration with neurosurgeons at UNM, we are developing techniques for the rigorous analysis of intracranial parameters, through careful application and tailoring of techniques in signal processing and system identification. Our goal is to identify potential biomarkers that may be useful as an early indicator of patient outcome. Accurate predictive capabilities may help guide earlier treatment for preventing secondary injuries.
Motor Control in Parkinson's Disease
Characterization of motor performance in Parkinson's disease can can help elucidate faulty feedback mechanisms in the brain. Linear dynamical systems have been shown to be effective models of manual pursuit tracking, and system parameters such as damping ratio and natural frequency potential biomarkers. High-fidelity characterization of motor processes, and correlation of these models with observable brain processes (e.g., through fMRI or EEG data) can provide insight into compensatory mechanisms in the brain in Parkinson's disease. This is joint work with Dr. Martin J. McKeown and Dr. Z. Jane Wang at UBC.
This material is based upon work supported by the National Science Foundation under Grant Numbers CNS-1329878, CMMI-1335038, CMMI-1254990. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.