- Research on "Cross-Embodiment Learning for Robot Manipulation using Egocentric Human Play data" in the Robot Learning and Reasoning (RL2) lab under Prof. Danfei Xu (Assistant Prof. and Research Scientist, NVIDIA AI)

Graduate Teaching Assistant - CS 6476 Computer Visionhttps://www.humphreyshi.com/vision

Developing perception and scene understanding for Autonomous Vehicles at Intersections.

Graduate Teaching Assistant for the course CS6476/4476: Computer Visionhttps://www.humphreyshi.com/vision

• Google Summer of Code'23 at Ivy - Unify.ai• Developed a multi-backend framework support (PyTorch, JAX, NumPy, Tensorflow) for GradSLAM, an end-to-end differentiable SLAM system, using graph compiler, enabling deployment with highly optimized frameworks like JAX.

• Scene Understanding for Autonomous Driving in Adverse weather (with Queensland University and ZF Friedrichshafen Group)• UAV-based Visual Remote Sensing for Automated Building Inspection (UVRSABI) - with IHub Project Mobility and Govt. of India • DodgeDrone(ICRA 2022) - Vision-based Agile Drone Flight.

• Improved Visual Odometry and SLAM pipelines for Level-5 Autonomy.• Semantic Segmentation of LiDAR scans built upon RangeNet++ andSuMa (Efficient LiDAR-based Semantic SLAM) on KITTI Dataset.• Proposed a semantic variant of ICP (Iterative Closest Point) algorithm for incrementally building maps from point cloud data(LiDAR) for Autonomous Driving. Improved performance over vanilla ICP by 97% (matching loss) and 50% (convergence time). Evaluated on Semantic KITTI Dataset and achieved real-time latency of 4 scans/sec. Libraries: Point Cloud Library(PCL).

Technical club of NIT Surat1. Asia-Pacific Robot Contest (RoboCon) 2019: • Led a 15-person team in Asia-Pacific Robot Contest (RoboCon) 2019, overseeing both technical and managerial aspects of building a 4-wheel Holonomic Drive and a Quadruped Robot. • Developed autonomous motion of holonomic drives using line following and wheel encoder-based odometry. • Developed a dual camera-based perception system interfaced on Raspberry Pi 3B for legged robot. • Secured 13th all over India in the final round at the National Robotics Contest held at IIT Delhi clearing 2 rounds of Design Details Document and Video Submission among over 100 university teams.2. Asia-Pacific Robot Contest (RoboCon) 2018: • Represented SVNIT at the Asia-Pacific Robot Contest (ABU RoboCon) 2018.• Our team of 20 went on to develop 2 Robots- an Autonomous 3-Wheel Omni Drive and a remote-teleoperated 3-Wheel Omni Drive which had to perform specific tasks in a given arena according to the problem statement.• Developed software for autonomous motion using Line Following on a 3-wheel omni-drive through feedback from sensors like LSA line sensor, Gyroscope, IMU and Encoders on Atmel AVR and ARM microcontroller hardware.• Developed teleoperation-based control of the omni-drive using a remote Playstation controller and prepared hardware circuitry using GCB (General Circuit Boards).• Achieved 12th rank overall in the competition held at the National Level among over 100 teams that participated in the competition.3. Organized INSIGHT 1.0, a technical symposium with a footfall of 500-plus people.4. Mentored Embedded Systems and Robotics projects such as RFID-based Identification and Wireless control of mobile robots.

• "Face Recognition Using Deep Learning" under Dr. K.P. Upla, Assistant Professor, ECED, NIT Surat.• Implemented the Facenet paper from Google and tested it on a custom dataset (of around 25 students).• Developed different modules like Face Detection, Alignment, and Recognition using the Inception CNN Model to extract facial features and classify them.