Computer Vision and Analytics Programmer

• Feb 2023 - Present

Engineering Intern, Self Driving, ADAS Team

• Jun 2023 - Sep 2023

◦ Sensor Fusion: Implemented factor graph-based vehicle state estimation using IMU and Wheel Odometry sensors to perform robust ego-pose estimation. ◦ Simulation: Performed data generation using Carla for validation of pose estimation algorithms. ◦ Evaluation: Evaluated and showcased the benefits of using Factor Graph against EKF for sensor fusion. ◦ Sensor Fusion: Implemented factor graph-based vehicle state estimation using IMU and Wheel Odometry sensors to perform robust ego-pose estimation. ◦ Simulation: Performed data generation using Carla for validation of pose estimation algorithms. ◦ Evaluation: Evaluated and showcased the benefits of using Factor Graph against EKF for sensor fusion.

Senior Systems Software Engineer

• Mar 2022 - Jun 2022

Implemented Checks and fixed bugs in the validation pipeline to ensure maps being created are correct and appropriate flags are raised for the errors detected. Implemented Checks and fixed bugs in the validation pipeline to ensure maps being created are correct and appropriate flags are raised for the errors detected.

Module Lead - Robotics

• Jan 2022 - Mar 2022

Robotics Software Engineeer

• Feb 2020 - Dec 2021

• Implemented 13-state Quaternion-based Error-State Extended Kalman filter to fuse odometry and pose estimates using fiducial marker. • Packaged a VIO algorithm in ROS and was responsible for testing and Validating the algorithm. • A Model-based Auxiliary odometry was developed for an emergency landing in case of primary Navigation Stack failure. • Semantic Segmentation networks were trained and tested for a localization algorithm. It was further optimized using… Show more • Implemented 13-state Quaternion-based Error-State Extended Kalman filter to fuse odometry and pose estimates using fiducial marker. • Packaged a VIO algorithm in ROS and was responsible for testing and Validating the algorithm. • A Model-based Auxiliary odometry was developed for an emergency landing in case of primary Navigation Stack failure. • Semantic Segmentation networks were trained and tested for a localization algorithm. It was further optimized using TensorRT to achieve about 4x performance boost over PyTorch implementation with marginal loss inaccuracy. • Using an off-the-shelf environment in Unreal Engine 4, object detection network was trained and tested in the simulation environment using AirSim. Moreover, SIL testing was done with Gazebo to validate the entire obstacle detection pipeline.

Research Assistant

• Apr 2019 - Jan 2020

•Developed a differentiable Line based Lidar-Camera extrinsic calibration method. Resection loss was used as cost function and was implemented using Kornia package. Rotation and Translation parameters were optimized using PyTorch’s Autograd package. •Segmentation of point cloud was performed by mapping 3D points to the segmented image using extrinsic parameters. Fast-SCNN network (Rudra PK et. al.) was used for image segmentation and was fine-tuned on Indian driving dataset (IDD)… Show more •Developed a differentiable Line based Lidar-Camera extrinsic calibration method. Resection loss was used as cost function and was implemented using Kornia package. Rotation and Translation parameters were optimized using PyTorch’s Autograd package. •Segmentation of point cloud was performed by mapping 3D points to the segmented image using extrinsic parameters. Fast-SCNN network (Rudra PK et. al.) was used for image segmentation and was fine-tuned on Indian driving dataset (IDD) to improve its accuracy on Indian driving conditions. •Configured LOAM(Ji Zhang et. al.) for Velodyne’s VLP-16 Lidar and used it as an odometry source for an Autonomous Vehicle project. To improve Navigation Stack’s robustness against highly dynamic environments (Case where LOAM fails), the odometry from LOAM was fused with GPS using an Extended Kalman filter. •Part of the core team to built a Hexacopter for pick and place operation by following GPS way-points which was exhibited during IIIT Hyderabad’s R&D showcase. Show less •Developed a differentiable Line based Lidar-Camera extrinsic calibration method. Resection loss was used as cost function and was implemented using Kornia package. Rotation and Translation parameters were optimized using PyTorch’s Autograd package. •Segmentation of point cloud was performed by mapping 3D points to the segmented image using extrinsic parameters. Fast-SCNN network (Rudra PK et. al.) was used for image segmentation and was fine-tuned on Indian driving dataset (IDD)… Show more •Developed a differentiable Line based Lidar-Camera extrinsic calibration method. Resection loss was used as cost function and was implemented using Kornia package. Rotation and Translation parameters were optimized using PyTorch’s Autograd package. •Segmentation of point cloud was performed by mapping 3D points to the segmented image using extrinsic parameters. Fast-SCNN network (Rudra PK et. al.) was used for image segmentation and was fine-tuned on Indian driving dataset (IDD) to improve its accuracy on Indian driving conditions. •Configured LOAM(Ji Zhang et. al.) for Velodyne’s VLP-16 Lidar and used it as an odometry source for an Autonomous Vehicle project. To improve Navigation Stack’s robustness against highly dynamic environments (Case where LOAM fails), the odometry from LOAM was fused with GPS using an Extended Kalman filter. •Part of the core team to built a Hexacopter for pick and place operation by following GPS way-points which was exhibited during IIIT Hyderabad’s R&D showcase. Show less

Member of the Technical Staff, Embedded Engineer

• Jun 2017 - Dec 2018

• Responsible for the firmware development of the inertially stabilized Pan Tilt Unit of a product called T-Rex (an advanced long range reconnaissance and observation system). • The system was made stable and robust to high velocity winds. It was validated in wind tunnel at 60kmph and reported good performance. • Worked on Soft/Hard Iron compensation for magnetometer to make attitude measurement more stable when placed near stepper motors or any other magnetic field generating… Show more • Responsible for the firmware development of the inertially stabilized Pan Tilt Unit of a product called T-Rex (an advanced long range reconnaissance and observation system). • The system was made stable and robust to high velocity winds. It was validated in wind tunnel at 60kmph and reported good performance. • Worked on Soft/Hard Iron compensation for magnetometer to make attitude measurement more stable when placed near stepper motors or any other magnetic field generating devices. • Implemented Cascaded PID loop to make system responsive, disturbance resistant and achieve smooth performance. • Investigated the accuracy and performance of Kalman filter, Median filter, IIR filter and moving average filter on sensor noise filtering. Show less • Responsible for the firmware development of the inertially stabilized Pan Tilt Unit of a product called T-Rex (an advanced long range reconnaissance and observation system). • The system was made stable and robust to high velocity winds. It was validated in wind tunnel at 60kmph and reported good performance. • Worked on Soft/Hard Iron compensation for magnetometer to make attitude measurement more stable when placed near stepper motors or any other magnetic field generating… Show more • Responsible for the firmware development of the inertially stabilized Pan Tilt Unit of a product called T-Rex (an advanced long range reconnaissance and observation system). • The system was made stable and robust to high velocity winds. It was validated in wind tunnel at 60kmph and reported good performance. • Worked on Soft/Hard Iron compensation for magnetometer to make attitude measurement more stable when placed near stepper motors or any other magnetic field generating devices. • Implemented Cascaded PID loop to make system responsive, disturbance resistant and achieve smooth performance. • Investigated the accuracy and performance of Kalman filter, Median filter, IIR filter and moving average filter on sensor noise filtering. Show less

Intern

• Jul 2016 - Dec 2016

• Contributed in the Development the firmware of Motion Control Unit of an inertially stabilized Pan-Tilt Unit which resulted in successful demo. • Using IoT Module, Created an interface to communicate with the system over Internet. • Contributed in the Development the firmware of Motion Control Unit of an inertially stabilized Pan-Tilt Unit which resulted in successful demo. • Using IoT Module, Created an interface to communicate with the system over Internet.

Summer Research Intern

• Jun 2016 - Jul 2016

Development of a notebook reader for the visually challenged readers. • A helmet like a device embedded with a camera. • The Device acts as a slave and uses laptop's computing power to process images and transfers the extracted text back to the device in order to recite it. • My role was to design and specify the hardware required for it. • Designed a converter board and a hub to which all the peripherals are attached. Development of a notebook reader for the visually challenged readers. • A helmet like a device embedded with a camera. • The Device acts as a slave and uses laptop's computing power to process images and transfers the extracted text back to the device in order to recite it. • My role was to design and specify the hardware required for it. • Designed a converter board and a hub to which all the peripherals are attached.

Intern

• May 2015 - Jul 2015

Study of Steam Turbines and its energy losses and hence calculating its efficiency. • To study the working of different parts in Steam Turbine. • Visits to power plant and understand the functioning of every unit in plant. Study of Steam Turbines and its energy losses and hence calculating its efficiency. • To study the working of different parts in Steam Turbine. • Visits to power plant and understand the functioning of every unit in plant.