Udit Singh Parihar

Computer Vision, DL and SLAM Research Engineer

Qualcomm XR Research

Biography

Hey! I am currently working as a Computer Vision, DL and SLAM Research Engineer at Qualcomm XR Research, where I design and optimize multi-camera Visual-Inertial SLAM systems (shipped in Google AndroidXR and Samsung Moohan). My work focuses on improving 6DoF accuracy, robustness, and power efficiency for AR/VR experiences. I have also contributed to 6 accepted patents spanning hybrid classical-DL feature tracking, VLM semantic priors for SLAM, and illumination-robust matching.

Previously, I was at OLA Electric, working on end-to-end autonomous driving agents, semantic BEV occupancy grids, and LiDAR-based mapping. Our team achieved a Silver Medal in the Kaggle Image Matching Challenge 2022, and our research on BEV appearance estimation was featured at ICRA.

I completed my Master’s at the Robotics Research Center, IIIT Hyderabad, advised by Prof. Madhava Krishna. My research focused on robust pose graph SLAM and rotation-invariant feature matching, with publications in ICRA, IROS, and VISAPP.

Download my resumé.

Interests

SLAM / State Estimation
Differentiable Computer Vision
Optimization
Robotics

Education

MS by Research in CSE with specialization in Robotics, 2021
International Institute of Information Technology, Hyderabad
B.Tech in Mechanical, 2018
Indian Institute of Technology Jodhpur

Experience

Computer Vision, DL and SLAM Research Engineer

Qualcomm

Jul 2023 – Present Bangalore, India

Designed architecture for multi-camera support in 6DoF Visual-Inertial SLAM; integrated multi-camera constraints into tracker and pose graph optimizer (shipped in Google AndroidXR & Samsung Moohan).
Contributed to 6 accepted patents in Deep Learning & SLAM, including hybrid classical-DL feature tracking, VLM semantic priors for SLAM perception, and illumination-robust matching.
Optimized Visual-Inertial Odometry by removing redundant IMU intrinsic calibration from the state vector and deriving analytical closed-form warping matrices for camera distortion, reducing power hotspots.
Built ground-truth infrastructure for 6DoF accuracy evaluation involving mechanically constrained setups and 360-spanning ChArUco rigs.

Computer Vision and SLAM Research Engineer

OLA Electric

Jul 2021 – Jul 2023 Bangalore

Developed an end to end autonomous driving agent using cameras, GPS and IMU sensors. Ported the agent from Carla simulator to NuScenes Dataset.
Converted the pytorch model to TensorRT and developed a ROS wrapper to run on real Mahindra E2O car achieving final control prediction at 25 HZ, in a zero shot paradigm.
Won the silver medal in the Kaggle Image Matching Challenge 2022 by developing an Ensemble of Deep feature matching algorithm of SuperGlue and LoFTR.
Extended the Lidar based mapping and localization LeGO-LOAM SLAM for the Velodyne and Ouster lidars and ported ROS1 to ROS2 in C++.
Trained Self Supervised Depth estimation PackNet-SfM on Indian driving dataset and on Carla simulator dataset.

Graduate Research Assistant

Robotics Research Center, IIITH

Aug 2018 – Jul 2021 Hyderabad

Worked on the intersection of SLAM, Computer Vision, Deep Learning, and Robotics. Developed robust pose graph constraints using scene semantics and developed rotation invariant deep feature descriptors for feature matching.
Published in ICRA and IROS conferences.

Featured Publications

Udit Singh Parihar, Aniket Gujarathi, Kinal Mehta, Satyajit Tourani, Sourav Garg, Michael Milford, K. Madhava Krishna

October 2020 In IROS 2021

RoRD: Rotation-Robust Descriptors and Orthographic Views for Local Feature Matching

We achieve high viewpoint invariance local feature matching using rotation homographies during training and converting perspective image to orthographic for feature matching.

PDF Code Project Video

Sai Shubodh Puligilla, Satyajit Tourani, Tushar Vaidya, Udit Singh Parihar, Ravi Kiran Sarvadevabhatla, K. Madhava Krishna

April 2018 In ICRA 2020

Topological Mapping for Manhattan-like Repetitive Environments

Used semantics understanding to assist loop closure detection and localization in SLAM framework in challenging warehouse environment. Incorporated topological edges in a pose graph optimization.

PDF Code Video