Sandeep Kota - Machine Learning Engineer | Computer Vision Engineer

About me

Hello! I'm Sandeep Kota, a passionate Computer Vision Engineer and Robotics specialist based in the United States. With a strong background in Machine Learning and Robotics, I'm dedicated to pushing the boundaries of autonomous systems and artificial intelligence.

My journey in the field of robotics and computer vision has led me to work on exciting projects, from conventional CV based trailer tracking systems for major automotive companies to developing autonomous drones for warehouse management. I thrive on challenges that require innovative solutions, particularly in the areas of deep learning, computer vision, and robotics.

With a Master's degree in Robotics from the University of Maryland College Park and hands-on experience of working for leading automotive company, Continental Automotive, I've honed my skills in C++, Python, ROS, and various machine learning frameworks.

When I'm not coding or working on robotics projects, you can find me exploring the latest advancements in AI and computer vision, or writing about technology. Currently exploring Large Language Models (LLM's), Vison Language Models (VLM's) and its applications in the field of robotics. Learning about finetuning, inference and efficient deployment strategies.

Explore My Work Experience

Education

University of Maryland, College Park

Master of Engineering in Robotics

Graduation Year: 2021

SASTRA University, Thanjavur

Bachelor of Technology in Mechatronics

Graduation Year: 2018

Tech Stack

Python

C++

PyTorch

TensorFlow

NumPy

Scikit
Learn

Pandas

CUDA

Weights &
Biases

Hugging
Face

ROS

Git

GitHub

Docker

AWS

Streamlit

Experience

My professional journey in the field of Computer Vision and Robotics

Software Engineer - Computer Vision

Continental Automotive

Oct 2021 - May 2024

Auburn Hills, MI - USA

Stellantis Project - Trailer Reverse Assist
- Improved performance of keypoint-tracking-based trailer tracking algorithm using geometric computer vision concepts, reducing trailer angle error KPI by 5%. The system learns trailer features during calibration/driving phase, then tracks them in a specific ROI during reversal maneuvers to calculate trailer angle.
- Implemented Kalman Filtering-based ROI adjustment to improve tracking in reverse direction by 10%, enabling a wider angle of operation ,greater than 60 degrees, of the feature on gooseneck and fifth-wheel trailers.
- Analyzed and resolved bugs reported from vehicle testing CAN signal data dumps. Extracted meaningful information and probable failure causes from thousands of rows of CSV data using Python, Pandas, and visualization libraries.
- Conducted research and prototype development for advanced deep learning-based solutions for next-generation products.
- Core stack used was C++ and Google Test (gtest) for production, targeting an ARM-based embedded processor. For analysis and prototyping, used Python with libraries including Pandas, NumPy, Matplotlib, Plotly, and PyTorch. Utilized Git and GitHub for version control.
Aurora Autonomous Trucks - Level 4 Fallback Path
- Developed a Convolutional Neural Networks(CNN) based trailer pose estimation proof-of-concept for fallback path planning, capable of SAE level 4 autonomy. The fallback system is designed to autonomously drive the truck to a safe location during main system failure. Utilized Python libraries including PyTorch, PyTorch Lightning for algorithm development, Weights & Biases for experiment tracking, and CVAT for data labeling.
- Engineered a high-performance data collection pipeline facilitating multi-terabyte data transfer between global development hubs.

Robotics Engineer

Infovision Inc.

Jun 2021 - Oct 2021

Richardson, TX - USA

Autonomous Drones for Warehouse Management Application
- Hands-on project involving building drone hardware, implementing visual inertial odometry in C++ and controls using ROS and PX4 autopilot.
- Optimized existing A* algorithm based path planning algorithm reducing planning time by 15%. Implemented stereo depth based dynamic obstacle avoidance using ROS and OpenCV.
- Spearheaded the implementation of version control by establishing a bitbucket repository enabling CI/CD pipeline for global development team.

Robotics Engineer

Flux Auto

Oct 2018 - May 2019

Bengaluru, India

Perception and Localization for Autonomous Trucks
- Development of a lighter U-Net based lane detection improving lane detection accuracy by 8%, with about 10% reduced params.
- Integrated hardware and software components to develop a ROS based perception system for autonomous trucks.
- Designed and implemented a comprehensive data collection pipeline for autonomous trucks, leveraging rosbag data structure to capture information.
- Conducted research and prototype development for advanced deep learning-based solutions for next-generation products.
- Developed and implemented an online calibration procedure to ensure precision and accuracy in data collection during drive scenarios.
- Core stack used was Python, Keras, ROS for algorithm development, Git and GitHub for version control.

Projects

Showcasing some of my key projects in robotics and AI

Robotics

ARIAC 2020

Agile Robotics for Industrial Automation (ARIAC) is a simulation based competition organized by the National Institute of Standards and Technology (NIST). The premise of ARIAC 2020 is to build kits of different types of products based on received orders from the system. To build this kits, a gantry robot with two UR10 arms attached to its torso and vacuum grippers can be moved around a warehouse to retrieve t he parts from each order. These parts can be found in shelves and bins. The robot picks up the parts and places them on a specified AGV. Once a kit is completed, the AGV is shipped to fulfill an order. ARIAC 2020 presents the following agility challenges include high priority order, faulty parts, faulty gripper, flipped parts, sensor blackout and moving obstacles. Some of the tools used for this competition include C++, ROS, Gazebo, MoveIt and ROS TF.