Bharath Kumar
Bharath Kumar

Bharath Kumar Ramesh Babu

Software Engineer II, Mapping — Aurora

Sunnyvale, California

I am a software engineer working on HD mapping for autonomous vehicles at Aurora. My work spans HD map generation, localization, deep learning pipelines, and 3D geometry algorithms for autonomous driving. My background includes perception, SLAM, and robot manipulation, with experience across autonomous vehicles, drones, and manipulators. I am proficient in C++ and Python, and passionate about building reliable, high-quality systems at the intersection of computer vision and robotics.

Experience

Work

Aurora — Software Engineer II, Mapping Jun 2023 – Present
Mountain View, CA
  • Developed HD map generation and localization workflows for Aurora's autonomous driving system. Maintained scalable HD map infrastructure and developed geometric and semantic validation algorithms to ensure high quality outputs.
  • Engineered automated pipelines leveraging deep learning and geometric algorithms to extract lane boundaries and traffic lights/signs from LiDAR and camera data.
Flytbase — Robotics Engineer Jul 2020 – Jun 2021
India
  • Developed perception and navigation modules for drone-based warehouse inventory product and helped progress it to deployment testing at 3 prominent warehouses in the USA.
  • Engineered barcode detection and sensor fusion pipelines. Worked on Mask-RCNN/MultiNet models and ORB-SLAM for aisle/bin segmentation and localization. Delivered RRT* planner module for drone surveillance product.

Education

Worcester Polytechnic Institute Aug 2021 – May 2023
Massachusetts, USA · CGPA: 4.0/4.0

M.S. Robotics Engineering

National Institute of Technology Tiruchirappalli Jul 2016 – Sep 2020
India · CGPA: 7.8/10

B.Tech. Instrumentation and Control Engineering · Minor in Computer Science Engineering

Publications

Projects

Open3D Contribution

Open3D — Open Source Contribution

Mar 2024 – Present

Landed pull requests enhancing mesh processing algorithms by adding UV map interpolation functionality and improving the TriangleMesh operator to handle UV maps more effectively.

Open Source 3D Vision
LiDAR Odometry

LiDAR Odometry Estimation using PointNet

Developed a learning-based odometry regression algorithm utilizing the KITTI lidar dataset. Built a PointNet-based architecture to filter dynamic objects and estimate 6D poses. Implemented feature matching in depth space, achieving results comparable to state-of-the-art while outperforming ICP.

SLAM Perception
Semantic Segmentation

Semantic Segmentation & Depth Estimation via Multi-Task Learning

Implemented and optimized an FCN-based MTL architecture achieving real-time inference at 12 FPS. Achieved 81% MeanIOU and 4m RMSE on CityScapes. Modularized the implementation to incorporate custom encoders/decoders.

Perception Deep Learning
Structure from Motion

Structure From Motion — Oxford Multi-View Dataset

Implemented ORB feature matching and Essential Matrix estimation for relative pose recovery. Built feature triangulation and bundle adjustment to construct a sparse 3D scene reconstruction.

3D Vision Perception
Pose Estimation AR

Augmented Reality & 3D Pose Estimation

Developed AR for 3D objects with perspective projection and homography. Implemented pose estimation using the Perspective-N-Point algorithm and improved performance with Kalman filtering, using a Rubik's cube as a fiducial.

3D Vision AR
Shadow Detection

Shadow Detection & Removal

Implemented an unsupervised segmentation algorithm for shadow/sub-region segmentation. Tailored a pipeline to match sub-regions and transfer illuminance based on a Gabor filter metric on the SBU shadow dataset.

Perception Image Processing
Laser SLAM Agricultural Robot

Laser SLAM for an Indoor Agricultural Robot

Developed a mobile robot that autonomously navigates indoor environments with laser SLAM using Kinect. Built plowing, sowing, and spraying mechanisms using Gmapping-based localization.

SLAM Robotics
Path Planning

Primer on Path Planning Algorithms

Implemented and compared graph-based and sampling-based motion planning algorithms including RRT, RRT*, Informed RRT*, PRM, D*, A*, and Dijkstra with a detailed comparative analysis.

Motion Planning Algorithms

Skills

Areas of Interest

Perception SLAM Motion Planning 3D Geometry Autonomous Vehicles Drones Robot Manipulators

Languages

C++ Python C Linux

Libraries & Tools

ROS / ROS2 OpenCV Open3D PyTorch CUDA Eigen Numpy Scikit-learn OpenGL

Courses

Computer Vision Deep Learning Motion Planning Robot Dynamics Data Structures & Algorithms

Research

Manipulation & Environmental Robotics Lab Jan 2022 – May 2023
Prof. Berk Calli · Worcester Polytechnic Institute & NERVE Center, UMass Lowell
  • Benchmarked grasp detection algorithms under various experimental conditions (Panda, UR5, RealSense, ZED 2i). Developed open-source software and simulator to benchmark grasping algorithms.
  • Implemented vision-based algorithms (GG-CNN, ResNet-based, Point Cloud Top Surface-based, Gripper Mask-based) for benchmark evaluation.
Vision, Intelligence & Systems Lab Aug 2022 – May 2023
Prof. Ziming Zhang · Worcester Polytechnic Institute
  • Developed a PointNet-inspired odometry regression model utilizing the KITTI lidar dataset. Built a two-stage architecture to filter dynamic objects and estimate 6D poses.
  • Achieved results comparable to state-of-the-art while outperforming ICP in inference time.

Internships

Brain Corp — Robotics Software Intern Aug 2022 – Dec 2022
San Diego, CA
  • Incorporated camera/lidar calibration module into the autonomy stack.
  • Developed an in-house performance and memory profiler for Brain Corp's SLAM system.
Aurora — Software Engineer Intern, Mapping May 2022 – Aug 2022
Pittsburgh, PA
  • Optimized lane boundary curve fitting in the HD Map creation pipeline, significantly boosting performance.
  • Developed visualization and metric evaluation tools for the curve fitting algorithm.
Flytbase — Robotics Intern May 2020 – Jul 2020
India
  • Created a simulator for drone-based warehouse inventory, extending the capabilities of Microsoft AirSim. Integrated the simulator into the codebase to facilitate testing of drone missions.