I study CS at Stanford on the AI track. I started in MatSci and MechE and moved to CS to apply math, ML, and prediction to hardware. I'm interested in autonomy and defense tech, and I play guitar.
Flies an indoor course with no GPS (maze, dark tunnel, window), lands and disarms on the pad, then re-arms by itself and flies back over the walls to the launch mark. Stereo visual odometry and a downward rangefinder fuse in PX4's EKF2, and a ROS 2 waypoint follower tracks the route. Built with Lucas Kniska and Petr Tupitsyn.
github.com/yashashaw/LongHarrierFinds gaps in low-altitude drone detection. Models the route a detection-avoiding drone would fly over terrain with a reverse Dijkstra flood and A*, maps where radar coverage is thin, and recommends sensor sites with minimax placement. Briefed to the Latvian Ministry of Defense and NATO. The production system is private. The repo is a public reimplementation on synthetic terrain.
The mission planner seen here (frontend) was vibe coded via Claude Code to use the preexisting backend that I had in order to do a visual presentation during my briefings.
github.com/yashashaw/public-hologramEnergy-optimal path planning for the SSI polar rover. Runs A* on a 10 m Antarctic elevation grid, costing each move for terrain, snow, surface roughness, wind, and cold, then refines the route on the real terrain. Prototyped around Concordia Station and checked against brute-force Dijkstra. Rover parameters are placeholders until the rover exists.
github.com/yashashaw/rover_pathfinder