Year-long ECE496Y1 capstone for the National Research Council of Canada's autonomous-rotorcraft program: a four-person team restoring and extending an autonomous Bell-412 flight stack in high-fidelity simulation. Co-designed and implemented the online path replanner in MATLAB: obstacle avoidance modeled as Dubins-style circular arcs, the intermediate velocities and accelerations derived, and quintic splines stitched into a dynamically feasible trajectory. Also owned environment integration and DevOps across two Linux setups: resolved dependency conflicts, cleaned up the NRC's initial ROS codebase and stood up the shared repo. The team's stack runs from perception to control (OpenCV/YOLO detection, LVI-SAM and OctoMap SLAM, ACADO/qpOASES NMPC) and demonstrated waypoint cruise, obstacle replan and safe landing end to end in a ~25,500 LoC C++ ROS/Gazebo system.
Active Vimy product R&D for flight-controller-agnostic GPS-denial support on small UAVs. Built a working demo that fuses vision, inertial, and lidar-derived state into MAVLink-compatible GPS/odometry outputs, then continued hardening the architecture with deterministic replay tooling, Docker builds, and license-clean dependency gates.
Built prototype/demo architecture for a proposed Vimy maritime-domain-awareness pipeline fusing heterogeneous sensor streams into explainable, policy-aware outputs. The concept maps to DND/CAF multi-modal AI needs: SAR, optical, AIS, weather/ice, legacy radar, and OSINT sources fused into situational-awareness outputs with transparency hooks rather than black-box answers.