Developing PTZ camera orientation estimation and image based smoke detection under Prof. Falko Kuester.• Developing methods to utilize hundreds of petabytes of images• Learning to estimate the location of the sun from images for Pan-Tilt-Zoom camera orientation estimation via using deep learning.• Training BEiT, a type of Vision Transformer, to distinguish smoke from clouds.• Compiling smoke segmentation dataset using AWS SageMaker and MTurk

Working with Prof. Henrik Christensen together with Dr. Allen Yang on improving subpixel resolution for image feature detection.

Yonder Dynamics is student organization founded in 2016 focused on developing autonomous rovers for the international University Rover Challenge.• Led software team since 2018, in charge of development of sensor fusion, path following, ob-stacle avoidance, MoveIt interface, and inverse kinematics.• Led development of novel outdoor Bayesian exploration method based on RRT-Star.• Enhanced RGBDSLAM’s ability to incorporate dynamic objects using motion segmentation.Technologies: ROS, rviz, MoveIt, Gazebo, Arduino, C++, PyTorch, Linux4Tegra

Performed research and mentoring masters students under Dr. Allen Yang.• Enhanced RGBDSLAM’s ability to incorporate dynamic objects using motion segmentation.• Trained a siamese neural network to perform loop closure detection for SLAM using ScanNet and Google Streetview data.• Developed state of the art image feature detection using deep learning for 3D reconstruction and pose estimation in collaboration with Prof. Henrik Christensen at UCSD. Extended abstractaccepted to ICCV2021 workshop. Inproceedings at CVIU.• Developing better object pose estimation for RGBD images.Technologies: Computer Vision, C++, Eigen, PyTorch, CUDA, Tensorflow, OpenCL, OpenCV, Point Cloud Library, Open3D, Kubernetes, Docker