Developed and applied statistical techniques to construct worst-case estimates of risk measures for robot policies in complex environments with analysis on the effects of distribution shift.Designed a formal analysis tool for deep neural networks, enabling rich interpretability by computing reachable sets, fixed points, regions of attraction, and certifying invertibility.

Course Assistant for: AA 277: Multi-Robot Control and Distributed Optimization AA 274A: Principles of Robotic Autonomy I AA 228: Decision Making Under Uncertainty AA 212: Advanced Feedback Control Design AA 151: Lightweight Structures Supported graduate students by creating course content, holding office hours, and grading assignments.Student feedback: - "Joe did a great job suiting his teaching to your background, explaining things clearly and concisely." - "Joe is very good at thinking through problems with you and giving insights that you wouldn't normally see." - "Joe gave very clear explanations on concepts, was very helpful answering questions, and was very patient when explaining particularly confusing topics."

Mentored nine high school students in machine learning projects spanning diverse topics from medical imaging to reinforcement learning for video games, with two students publishing in high school journals.Student feedback:- "Joe was very resourceful and knowledgeable. Very patient and was willing to teach and help along the way."- "Joe was always willing to explain and guide me through each step and help me whenever something confusing came up. His instructions were clear, and he was very supportive."

Evaluated the generalization of diffusion-based robots in unseen environments by estimating the cumulative distribution of performance with rigorous uncertainty quantification.Created binomial_cis, the first Python package for computing optimal binomial confidence intervals.

Extended research on out-of-distribution generalization to accurately predict performance of YOLOv5 drone detection networks on unseen datasets.Curated the DetFly drone detection dataset on Roboflow, leading to a more robust data analysis pipeline, improving ease of data analysis, data augmentation, training, and out-of-distribution testing.

Surveyed requirements, realities, and opportunities for using GPS signals in autonomous vehicle navigation.Documented and communicated results with three offices within U.S. Department of Transportation.

Collaborated on a three-person team to design a Cooperative Adaptive Cruise Control platooning algorithm.Led the stability analysis for the platooning algorithm, ensuring damping of traffic disturbances which was validated on FHWA’s CARMA vehicles in closed course tests at highway speeds.

Performed feasibility studies for synthetic aperture remote sensing of ice sheets using unmanned aircraft.

Helped customers, trained new hires, resolved day-to-day challenges.