Vision, Intelligence and Systems LabAdvisor: Prof. Ziming ZhangThe primary goal of my research involved improving the performance of VRCNet, a deep neural network for point cloud completion of real-world scanned incomplete point clouds due to occlusion and noise.Tech Stack: CUDA, Python, PyTorch• Conducted an ablation study to enhance point cloud completion using VRCNet model by experimenting with varied learning rates, advanced optimizers, and new activation functions.• Achieved a notable 12.5% accuracy improvement over the baseline performance.

Tech Stack: ROS, Gazebo, Python, MoveIt, RViz, Git• Engineered a simulation-hardware interface for a 6 DOF robotic arm for pick and place operation.• Utilized ROS, Gazebo and MoveIt to integrate software and hardware.

Tech Stack: CUDA, AWS, Git, Python, TensorRT• Trained object detection and classification models for automated waste segregation by a gantry robot.• Used YOLOv5 and ResNet architectures and deployed them on Jetson Xavier with TensorRT and CUDA for real-time waste management processing, utilizing AWS S3 for effective data pipeline control.• Attained over 90% detection accuracy on the most commonly encountered waste items.