Contributed to development of the autonomy stack for a quadcopter, with primary contributions in trajectory generation, path planning, 3D mapping, and learning-based depth estimation.

• Augmented quadcopter’s autonomy stack with trajectory generation capable of obstacle avoidance and breadth-first search for safe start and goal positions, increasing safety and reliability while flying

• Utilized runtime polymorphism to modularize navigation pipeline, improving code abstraction and allowing for multiple flight modes with unique implementations

• Evaluated stereo depth DNNs on Nvidia Jetson; calibrated Arducam sensors for disparity to depth conversion, wrote nodes to run TensorRT and ONNX models, and explored quantization options using MATLAB and Nvidia TAO

• Integrated multi-sensor OctoMap and Voxblox 3D mapping algorithms into a C++ ROS package, enabling quadcopter to map occupancy by combining input from forward- and downward-facing RealSense cameras

Worked on an automotive situational awareness system that used object detection and tracking on nearby vehicles and pedestrians to anticipate risky scenarios.

• Projected tracked objects from 2D image into 3D world to evaluate risk posed to system user

• Implemented a Kalman filter node in ROS to smooth tracked object trajectories

• Created an application with Nvidia Isaac SDK to test detection algorithms in simulation

Augmented welding robot recovery mechanism by allowing it to recalibrate in any reachable end effector orientation/position.

• Utilized proprietary programming language KAREL to implement solution

• Tested and validated implementation in ROBOGUIDE simulation, as well as on the robot

Assisted in the characterization of electrohydrodynamic printing dynamics for a novel dual-barrel configuration, which has potential applications in printed electronics and bioengineering.