• Implemented BiseNet, PIDNet models for segmentation, achieving 156fps inference speed with mean IoU of 96%
• Accelerated model inference by 40% using TensorRT and deployed on edge devices (Jetson Nano)
• Implemented LiDAR based obstacle avoidance and autonomous navigation algorithm
• Wrote production level robotic arm code for pose estimation and control based on weld plates detection
• Trained the U-Net model for semantic segmentation achieving a mean IOU of 94%
• Labelled and compiled the welding joints dataset, built custom preprocessing pipelines for DNN model input
• Modelled a custom mount using CAD software and 3D printed it to securely hold the welding torch and vision system