Research

NIMBUS lab UAVMy research brings together aerospace engineering, control theory, and computer science to fully integrate high-level vehicle intelligence with low-level vehicle control by co-designing the holistic robotic system. Historically, the design of small aerospace vehicles has been the almost exclusive domain of aerospace engineers, but this is quickly changing. Theorists and engineers have recognized the flexibility and intelligence that can come from computing activities onboard these vehicles. My broad background in various engineering and computer science disciplines has prepared me to examine and leverage what these traditionally disparate fields have to offer and make fundamental progress toward multidisciplinary co-design of robotic vehicles.

The separation of engineering and system design into separate, but overlapping fields of study has yielded tremendous results in modern technology. However, in some cases (e.g. control theory and artificial intelligence) these fields become sufficiently segregated that modeling, design, and development techniques overlap and only the contextual details remain different. This can lead to sub-optimal, brittle designs with undesirable emergent behavior when compositionality and composability assumptions are violated. This becomes increasingly problematic in long-duration vehicle deployment when an operator is unavailable to reset, fix, fuel, or otherwise attend to the vehicle's needs. Drawing inspiration from the robust symbiosis of thinking and acting in humans, co-design methodologies that interconnect and exploit the latest computing, control, optimization, and robot architecture research can lead to more robust and survivable robots.

To advance research in interdisciplinary robot co-design I have three primary research thrusts: 1) at the lowest reactive level, computing tasks can be co-regulated alongside physical system parameters in a closed-loop fashion. This yields robots which are more responsive to external stimuli without the need to engage in lengthy computing to make decisions. This is similar to the “fight or flight” response humans exhibit and is invaluable for survival. 2) At the higher deliberation/planning layer optimal decision making is generally limited to motion and task planning which don't consider the computation required to calculate the decision. This can infuse awareness into the intelligence of the robot by designing plans to be aware of the computing and physical processes and deliberating on those in addition to its actions, tasks, and motion plan. This is reminiscent of the higher-level deliberation that humans use in which they not only deliberate on the physical world, but also deliberate on their “mind” or emotions and thoughts. 3) Finally, any intelligent system (biological or not) requires monitoring the internal processes to determine how best to have its needs met. In humans this may be the immune system monitoring health markers and responding when it detects disease. For a robot, run-time monitoring of key system parameters, the run-time verification of new and novel plans, and the detection of failures is imperative to enhance long-duration deployment. Advancement in these key areas will create more reliable, energy efficient, and robust aerospace vehicles.