Designed and implemented an event driven state machine sub-system for real-time system replay, utilizing the DDS publisher-subscriber messaging service for the localization sub-system. This tool was instrumental in boosting the 3D point set registration workflow performance by 16%,

Developed a replay rendering 3D graphics application, utilizing Panda3d (OpenGL backend) and scene graphs, to render virtual endoluminal views and generate depth maps across lung anatomies. This application served as the foundation for deep learning computer vision-based localization, point cloud generation, mesh alignment using the produced depth maps, fly-by 3D registration and a tool for evaluating navigation performance in the field for the engineering and clinical teams.

Designed and implemented an extensible architecture using the abstract factory and decorator patterns. Facilitated the creation of 5+ metrics for automatic scope localization performance report generation. This system monitors performance across various release versions and is evaluated on a continuously evolving procedure dataset.

Optimized user experience by implementing a series of UI enhancements, streamlining workflows and screen state transitions.

Developed comprehensive test plans and associate software tool, to analyze and quantify sources of electromagnetic interference that could impact localization performance.

Conducted comprehensive V&V processes and dFMEA analysis for the localization and tracking subsystems, executing test cases and generating test reports for 510k submission and subsequent FDA approval of the Galaxy robotic system.

Research involved exploring the explainability of vision transformer and data efficient image transformer architectures for Visual Recognition.

Reverse Engineering of Multiscale, Pyramid Vision transformer and different flavors of vision transformer architectures to understand the Design philosophy behind the architecture design and understand information propagation through the architecture.

Researched on Autoregressive Moving Average networks for the task of dense predictions and object detection. Integrated the architecture with DeepLabv3/v3+ and Faster-RCNN as a feature extractor backbone and determined its adversarial robustness to input perturbations with Resnet18/34/50, as baseline networks on Pascal VOC.

Developed the application code for the Flash Memory Management system for our Electric Vehicle ECU in order to store the essential data of our telematics unit. Using USART, serially transmitted the stored data from the flash and onto the serial terminal to check data integrity. The code was customized to input and store data of any length in the flash(Circular DMA for retrieval and Queue's to input data).

Developed the functions required to enable the SPI peripheral for communication between the microcontroller and the IMU and wrote the application code to extract the linear acceleration and angular acceleration data for the ECU in the Interrupt and Non-interrupt Mode.

Developed the application code to extract the quaternions for the IMU which was essential to calculate the roll, pitch and yaw (Euler angles) of the system and calibrated the sensor to attain the equillibrium position.

Testing of the accelerometer, gyroscope and gps module by mounting it onto our EV for the purpose of datalogging of the variables(acceleration, angular acceleration ,quaternions and location).

Enabled and applied the tap detection functionality of STM32L4 to generate an interrupt for device wakeup and tested it on the electric vehicle by setting various thresholds and shock-time windows. This was used in implementing the software for intruder detection.