Welcome!

My name is Saptadeep Debnath, and I'm a Robotics Engineer with a passion for innovation and problem solving. With a Master's degree in Electrical and Computer Engineering, specializing in Robotics, from the University of Michigan, I have extensive experience in designing and developing advanced robotic systems.

As a Robotics Engineer at Equipment Technologies, Inc., I lead the Vision-Based Advanced Driver Assisted System initiative as the Product Owner, driving product innovation and improving operational efficiency. I designed an advanced CNN-based semantic segmentation network that precisely predicts crop rows for CAN-linked machine steering, delivering precision and accuracy. I built a state-of-the-art ROS architecture pipeline from scratch, ensuring seamless message relay from the prediction software to the steering control manager, optimizing operations and minimizing downtime.

As part of my undergraduate thesis at Fulda University of Applied Sciences, I explored LSTM network performance by adjusting datasets and built a ROS pipeline to control a robot with 98% accuracy via real-time freehand gestures.

My concentration areas include Robotic System Design, Machine Vision, Deep Learning, and Control Systems. I'm proficient in programming languages such as Python, C, C++, Bash and HTML. I also have experience using tools and technologies such as the Robotic Operating System (ROS), OpenCV, PyTorch, NVIDIA Jetson (Xavier AGX), and Machine Vision Cameras (Intel Realsense, ZED2i).

Please take a moment to explore the different tabs on my portfolio website and check out some of my highlighted projects. If you have any questions or would like to reach out to me, please don't hesitate to ping me at saptadeb@umich.edu

Resume

Quicklink to Projects

Object Tracking for Safety

This project demonstrates my ability to create practical solutions for real-world challenges through the development of an innovative method to detect moving objects and measure their distance from the camera with precision. The solution utilizes advanced object tracking techniques to accurately estimate object depth and issue proximity warnings. To achieve dependable results, the depth estimator method I developed incorporates cutting-edge scaling algorithms, resulting in precise measurements. This technology has a wide range of potential applications, including aiding in navigation for autonomous vehicles and robots as well as assisting humans in hazardous environments. My expertise in object tracking is showcased through this project, highlighting my capabilities to deliver effective solutions for complex problems.

Mobile Inverted Pendulum System

This project entailed the conceptualization, design, and realization of a two-wheeled Mobile Inverted Pendulum system. Employing a cascaded control architecture, the robot was stabilized, and operated manually with the aid of a DSM remote, as well as autonomously following pre-defined trajectories.

Open in GitHub

6-DOF Serial Link Robotic Manipulator

This project entails the design of a manipulator gripper alongside the development of a Python-based codebase for autonomous control of the serially connected motors. The project also integrates the use of a Kinect camera suite consisting of a 3D depth sensor and an RGB camera.

Open in GitHub

Slam and Path Planning implementation on MBot

This particular project was intended to be deployed on MBot, a robot designed in-house at the University of Michigan for educational purposes. However, as a result of the COVID-19 pandemic, the implementation was carried out in a simulated PyGame environment.

Open in GitHub

Invariant Extended Kalman Filtering for Robot Localization using IMU and GPS

The objective of this project is to deploy an In-EKF (Invariant-Extended Kalman Filter) based localization system and conduct a comparative analysis against an Extended Kalman Filter based localization system and a GPS-only dataset. The study will leverage the UM North Campus Long-Term Vision and LIDAR dataset, which is a comprehensive autonomy dataset collected for robotics research on the University of Michigan North Campus. The dataset encompasses data from various sensors, including planar lidar, omnidirectional camera, IMU, and GPS, to enable a detailed analysis.

Hand Gesture Control of a Robot using Intelligent Techniques

I had the opportunity to work under the supervision of Prof. Dr. Alexander Gepperth at Hochschule Fulda, Fulda, Germany, in collaboration with Dr. V. Kalaichelvi, to develop a system that translates free hand gestures to motion instructions on a TurtleBot running on Intel Atom. This involved implementing the recognition of free hand gestures using LSTM networks, utilizing TensorFlow with C++. Following this, the gestures were transformed into their respective pre-defined motions and parsed as motion instructions via ROS messages to the TurtleBot.

Non-Linear Modelling and Simulation of Unmanned Aerial Vehicle

Assisted Dr. Mary Lourde in developing different mechanical models for nonlinear mathematical modeling of a UAV for the purpose of simulation in MATLAB-Simulink environment. A PID controller was designed to control the attitude and position of the nonlinear model of the UAV, to achieve optimal performance.

Final Report

Indoor Localization of an Unmanned Aerial Vehicle

Collaborated with Dr. Jagadish Nayak on the design and development of a UAV that is spatially aware in an indoor environment devoid of GPS and SLAM algorithms. The primary focus was on sensor fusion by utilizing data from LIDAR Rangefinder and Optical Flow for accurate indoor localization.

Final Report