Yashas Prabhu

Benchmarked classical (DenseVLAD), deep-learning (NetVLAD), and foundation (AnyLoc, DINOv2) models to deploy a Visual Place Recognition architecture achieving 0.85 Recall@1 / 0.90 F1, exposed as a custom LangChain tool giving the agent spatial awareness. Generated a 4,000+ image synthetic dataset in NVIDIA Isaac Sim and Omniverse Replicator using multi-factor domain randomization (lighting, camera viewpoint, occlusion) to simulate non-deterministic warehouse environments.

Engineered a ROS2-PX4 integration architecture bridging the companion computer and sensors on aerial platforms to enable autonomous navigation. Developed custom ROS2 nodes to process and publish real-time obstacle data, feeding the drone’s collision-avoidance algorithms. Reduced sensor noise and improved obstacle-detection reliability in hardware flight tests by implementing moving-average filters in the LiDAR point-cloud pipeline.

Worked on the development of a research aerial vehicle platform. Worked on the development of a drone for indoor navigation in a GPS-denied environment using point LiDaRs. Worked on drone-swarming technology

Process Engineering Trainee under Technical Engineering Function Department.

Design and development of drones for search and rescue, crowd control, education, FPV shooting, and racing.Educated students about the subsystems of drones and its applications

Programming,CAD, FEA, Vibrations,Thermodynamics,Material science, Automotive, Aeronautics,Electronics