Ayushman Choudhuri

- Own QualiCam, a ROS 2 and NVIDIA Jetson real-time optical perception system for automated quality assurance, and carry it from architecture through integration, commissioning, and field deployment. - Lead deployment and maintainance of production ML models at customer sites, diagnosing and resolving software and integration issues both remotely and on-site to keep systems stable and efficient. - Improve the ROS 2 object detection (YOLO based) and tracking pipelines and custom vision algorithms for the F

- Master Thesis on the topic - "Explainable Transformer-based 3D Object Detection in LiDAR Point Clouds". - Researched and developed an attention map based approach to enhance the explainability of real-time 3D object detection using transformer-based models. - Built a ROS2-integrated pipeline to process LiDAR point clouds and visualize both 3D detections and saliency maps, enabling deeper insights into model inference and decision-making in autonomous systems. - Optimized the custom transformer based mo

- Built a proof-of-concept for QualiCam, a standalone harvest quality assurance system, and validated it at customer farms to assess crop quality under real-world conditions. - Built quantization and deployment pipelines for YOLO-based object detection models using TensorRT on the NVIDIA Jetson Orin NX platform.

- Spearheaded the development of a stereo vision-based 3D object detection and tracking pipeline for the Harvey autonomous harvester, built on ROS and deployed on edge hardware. - Showcased the 3D detection and tracking solution to an international audience at Agritechnica 2023, the world's largest agricultural technology exhibition. - Accelerated inference of state-of-the-art object detection models using OpenVINO, HailoRT, ONNX, and TensorFlow, implementing asynchronous pipelines to meet real-time per

- Development of the perception pipeline for machine learning, and computer vision applications to enable a mobile robot to perform inspection tasks in a construction site. - Designed and deployed a closed-loop LIDAR tilt system to increase the vertical field of view by 100% for close-range applications and developed the safety critical tilt control ROS package using C++ and deployed it on a NVIDIA Jetson Xavier AGX. - Worked with LIDAR Point Cloud Compression packages such as draco and algorithms such as

- Developed and implemented a machine-learning based calibration algorithm (C++) on a custom edge device for dynamic payload estimation of a self loading cement mixer truck. - Achieved a weight estimation accuracy of 98.5% with a maximum payload of 800Kg in dynamic conditions.

- Designed and developed control software (C/C++) and a robotic test bed for precise liquid dispensing for composite manufacturing at the Department of Aerospace Engineering. - Worked on the development of the motion control software for the 3 axis robotic gantry robot used in the robotic test bed.

- Worked on a AS/RS robotic system design for automating processes in medium and large scale warehouses. - Developed the chassis and frame design for a pick and place autonomous rover module for medium scale warehouses.

- Strange Matter Robotics was formed with the aim of developing reliable and robust gantry robots that would cater to the needs of the Indian manufacturing hubs, automate the MSME sectors and also bridge the gap between low-end and high-end industrial CNC milling machines. - Modulus 1 and Modulus R were 2 Gantry Robots that were developed and tested. Modulus X is a concept model that was developed to cater to the research and prototyping labs.

Worked on the product and mechanical design of a gantry based CNC mill "Modulus 1". Modularity and reliability of the product were of paramount importance. The project was carried out under the guidance of Prof Pankaj Upadhyay at the Department of Design, IIT Guwahati.

Robomanipal is the official robotics team of Manipal University . I had worked on the structural design and pneumatic circuit design of a badminton-playing robot which represented MIT at Robocon 2015 held at Pune. This year I have worked on the structural design of a pole climbing robot which participated at Robocon 2016 held at Pune . This year our team achieved an all India rank of 19 and also qualified for superleagues.

I worked on building ontologies and data collection in the field of mechanical assembly of aircraft. Protege 5.0 which is a ontology editor developed by Stanford University was used in building these ontologies . The project was under the guidance of Professor B Gurumoorthy , CPDM , IISc .

Worked on design of vertical pressure vessels as per ASME, API and NACE standards. Analysed and studies various damage mechanisms related to pressure vessels being used in the refinery.