Udit Agarwal

Summary

Developing Al-driven AMRs (Autonomous Mobile Robots), L2 Autonomous vehicle for reducing human intervention in shop floors, warehouses, and manufacturing plants. Working on deep learning architectures for V-SLAM, structure from motion, scene reconstruction, and monocular depth estimation. Worked on multiple sensor fusion (camera, lidar, GPS & IMU) for mapping and localization of autonomous vehicles, based on autoware. Finetuned YOLOv8 model on real world data captured from Intet RealSense Depth Camera, for robust pallet detection. Implemented detection and pose estimation from detected bounding boxes using docker container on Nvidia Jetson Nano. Performed multi-view object detection by minimizing epipolar error using Hungarian matching.

Summary

Extracted key information from document images for visual question answering. Experimented with various transformer-based architectures for scene and text understanding like ViBERT, VILBERT, Chargrid, etc. Performed multi-modal learning for VQA (visual question answering) using layout aware scene text transformers.