Robotics Engineer

Techinical Skills: Python, C++, PyTorch, JAX, ROS, TensorRT

Education

• M.S., ECE, Robotics specialization

  University of California, San Diego (Dec 2023)

• B.E., Mechanical

  BITS Pilani, Hyderabad Campus (July 2017)

Work Experience

Engineering Intern @ Rivian

Self Driving, ADAS Team (June 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.

Computer Vision and Analytics Programmer @ Qualcomm Institute - UC San Diego Division of Calit2, Dronelab (Jan 2023 - June 2023)

• 3D reconstruction

  Worked on using NeRF to reconstruct the scene and compared it’s accuracy with colmap and Artec Leo 3D scanner.

• NeRF	Developed a unique NeRF model to perform 3D reconstruction for layered (sliced) models eg. 3D printing

• Human Pose Estimation

  Developing a modular IMU-based human pose tracking solution to perform gait analysis of people with transtibial prostheses. This will help to validate the efficacy of prostheses being used.

Senior Systems Software Engineer @ Nvidia Corporation, Mapping Team, Pune, India (March 2022 - June 2022)

Implemented Checks and fixed bugs in the validation pipeline to ensure maps being created are correct and appropriate flags are raised if any issue is found.

Robotics Software Engineer @ Vimaan, Autonomy Team (Feb 2020 - March 2022)

• Sensor Fusion

  Implemented 13-state Quaternion-based Error-State Extended Kalman filter to fuse odometry and pose estimates using fiducial marker for Quadrotors.

• Navigation

  Proposed and completed POC for unique semantic-based Visual localization Algorithms that output pose (err- upto 10cm) in global reference frame resulting in drift-free localization of an Autonomous Drone.

• Machine Learning

  Trained and Optimized supervised Semantic Segmentation network using TensorRT to achieve about 4x performance boost on Jetson platform with marginal loss in accuracy.

• Obstacle Avoidance and Simulation

  Trained and Tested object detection network using AirSim + Unreal Engine 4. Moreover, SIL testing was done with Gazebo to validate the entire obstacle detection pipeline.

Research Assistant @ IIIT Hyderabad, Robotics Research Center, Hyderabad, India(Jan 2019 - March 2020)

• Lidar-Camera Calibration

  Developed a differentiable Line-based Lidar-Camera extrinsic calibration method. Github Project Link.

• Lidar SLAM

  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.

Projects

Tightly Coupled Visual-Inertial SLAM

Github Project Link

Summary:

EKF-based Visual Inertial Simultaneous Localization And Mapping (SLAM) is implemented using IMU data as odometry and a stereo camera as our measurement sensor. It is a popular algorithm where the robot trajectory and the mapping of the landmarks is done simultaneously. The landmarks here are essentially point-like features that are scale-invariant and rotation-invariant with favourable geometric characteristics (for eg corner). These features could be coming from any detectors and must be post processed to remove outliers. The accuracy of this Algorithm will get impacted if outliers are present.

Demo:

https://user-images.githubusercontent.com/20353960/235547305-e9a99c28-2272-4d79-a80b-6af542087867.mp4

Pose graph optimization

Github Project Link

Summary:

Implemented the back-end of Graph Slam in python from scratch, that optimizes the odometry generated by using constraints from other sources like loop closure or GPS.

Demo:

https://user-images.githubusercontent.com/20353960/141079035-9f608603-7cac-4fa2-b209-a68a4b141ac8.gif

Vibration Reduction using Input Shaping

Github Project Link

Input shaping can reduced the vibration in flexible structures (such as beams) while performing point-point motion and hence enabling the use of light weight materials in robotics.

Eulerian Motion and Color Magnification

Project Link

Eulerian Video Magnification can amplify tiny motion and color changes. I implemented it from scratch in python and demonstrated color changes of my face due to blood flow.