This project implements a real‑time engine knocking detection system on a resource‑constrained embedded platform. On‑board microphones capture combustion noise, a compact ML model classifies “knock” events in‑situ, and summarized diagnostics are pushed securely to a cloud backend for visualization, alerts and analytics. By enabling predictive maintenance and continuous performance monitoring, it helps prevent engine damage and optimize operational uptime.

Neurological disorders impact a large percentage of the global population and are a vast area of research in the clinical field. However, state of the art diagnostic measures such as MRI and CT scans are invasive and expensive. Saccades, rapid fixations in eye movements, are a promising but underutilized non-invasive biomarker for neurological abnormalities due to limited publicly available data and privacy concerns. To address this, we developed a pose-guided video generation model that produces synthetic saccades of three types: normal, bilateral hypermetria, and bilateral hypometria that mimic real eye movement patterns observed in clinical settings. We trained an MViT-V2 video classification model on the synthetic data as a baseline and tested its performance on clinical saccade data. Our approach demonstrates the potential of synthetic data to enable accurate and scalable saccade-based diagnostics, reducing the dependency on invasive imaging.

Our project focuses on optimizing baseball batting lineups using a novel statistic we created called Baserunner-dependent Run Production (BRP). Traditional lineup evaluations often overlook how the performance of one batter affects others in sequence. To address this, we analyze all possible 4-batter groupings (4-tuples) within a lineup and assign each a BRP value, which captures the expected run contribution based on player-specific probabilities and base-out states. We then use these values to evaluate complete 9-player lineups and select the one that maximizes total BRP. This approach shifts the focus from individual metrics like batting average or RBI to a more holistic, interaction-aware model. Our optimizer accurately identifies the most productive batting orders and can be applied to real player data to help teams make more strategic decisions. Initial results are promising, and we plan to further refine the model through pilot testing and performance validation.

In this project, we further developed our robust leader-follower protocol which autonomously coordinates a group, or swarm, of devices. The system is designed to seamlessly adapt to devices dropping out of the swarm unexpectedly and to any new devices joining the network. We focused on two main tasks this year: formal verification of the protocol and developing a demonstration with mobile robots. The formal verification proved that our protocol satisfies both safety and adaptability requirements.

At Design Day, we will have an interactive simulation, which shows how our protocol can coordinate up to 50 robots. We will also present a video of our robot demonstration, which uses five TurboPi robots as devices which autonomously work to each perform a task within different quadrants of a large grid.