Generative AI: Synthetic data generation using Diffusion Probabilistic Models.3D Blind Surface Reconstruction via Tactile SensingSynthetic Data in VR and AR (in collaboration with Aeronautical Development Agency (ADA) under Department of Defence Research and Development (DR&D) of India's Ministry of Defence)

Research on Generative Adversarial Networks (GANs) for preparation of synthetic / fake images facilitating data augmentation for training in object detection.Under the guidance of Prof. Pradipta Biswas and PhD Scholar Mr. Abhishek at the Intelligent Inclusive Interaction Design (I3D) Laboratory.- Utilized GANs as a technique for Data Augmentation, successfully generating 10,000+ synthetic images of metal specimens, resulting in a 40% increase in classification accuracy and 50% increase in training data diversity.- Curated a custom dataset of 5000 images in controlled light intensity and background environment settings.- Trained unpaired I2I (CycleGAN, UNIT) models outperforming paired models (Pix2Pix, BicycleGAN), boosting output quality by 4 folds. Evaluated these models based on diverse metrics for model selection in various applications.- Improved quantitative metrics, FID and Inception Score by 9% utilizing a custom transfer learned InceptionV3 model.

At Guidance, Control, and Decision Systems Lab (GCDSL) and Mobile Robotics Lab (MRL) in the Aerospace Department at IISc under the guidance of Prof. Debasish Ghose and Onkar Chopra.- Developed C++ implementatins using OOPs of weighted and unweighted versions of A*, Dijkistra’s, RRT and BFS path planning algorithms with a focus on memory optimization and code scalabiltiy. Utilized gprof for code profiling.- Optimized A*'s memory usage (Resident Set Size) by 15% compared to RRT and execution time by 200% than existing.- Enhanced code reusability by 60%, cutting development time by 40% through effective use of inheritance.- Inherited essential software architecture concepts to make software modular, scalable, efficient and optimal.- Corridor planning for Drone swarms.

* Designed and developed a full-stack mission planner interface for real-time monitoring of sensor telemetry and seamless mission deployment to an Autonomous Underwater Vehicle. * Developed an underwater buoyancy simulation for testing.under water robot's localization and path planning.* Configured sensor fusion for precise local and global position estimation and tracking of robot during missions.* Developed GPS-based auto-navigation system for AUV, converts GPS coordinates to UTM and local map coordinates, and sends the output to local and global planners and costmaps for path planning, with continuous feedback and result.* Developed ROS action server and client for fetching data from AUV to Ground Control Station (GCS) and sending mission commands back to AUV.

* Maintained team's budget and cash flow for financial year 2022-23.* Developed team's financial planning and assessment for International competitions during 2022-23.

* Designed and developed, Mars Rovers implementing robot localization and mapping for autonomous navigation.* During my tenure, team Shunya achieved impressive rankings in international competitions, including top positions in prestigious rover challenges like the European Rover Challenge (ERC) and University Rover Challenge (URC).

* Developed prototypes for testing robot localization using sensor fusion from IMU sensor and wheel encoders.* Developed and test code GPS coordinates and global pose estimation in ROS.* Developed 4-wheeled differential drive remote controlled ground robot, controlled wireless through ROS, with IMU sensor for path generation and L298N motor driver.* Interfaced wireless video feed from robot to ground control station through generic web-cams sensors over ROS master connection.* Explored and worked with Raspberry Pi, Jetson single board computers and Arduino, Teensy micro-controller boards, experimented with servo control, wireless Bluetooth data send and receive over ROS.

Designed and developed a full-stack interface for collaborative additive manufacturing utilizing two robotic arms.* Developed front-end in pyQt, dual robot connect, current joints position information along with, connection override safety, Gcode parsing, and percentage print display bar.* Developed back-end in python, utilized threading library for multi robot control simultaneously, used object-oriented programming for better code organization, flexibility and scalability.

Developed simulations for testing autonomous navigation for self driving on Autoware.ai, an open source software.* Tested simulations in ROS Gazebo environments.* Used CUDA for advanced graphics support and rendering of environment scenes, providing better data for training.