Nitish Gupta

Student • Engineer • Researcher

❝ We are getting close to intelligent machines, but they are showing the necessary weakness of intelligence. ❞ — Richard Feynman

In A Nutshell

With over four years of industry experience and a Master's in Computer Engineering, my research area revolves around developing and optimizing intelligent robotics systems such as autonomous vehicles or industrial robot manipulators. In particular, I am interested in algorithms for perception, planning, and control using tools such as ROS 1/2. Currently exploring reinforcement learning and real-time systems at RTIS-UCF Lab. Previously Research Engineer II at NHK International Corporation, USA, USA and Cooperative Autonomous Vehicle Researcher at CAVREL, UCF. I love to do gardening, play chess, read books/blogs, and travel in my leisure time.
Open to collaborations.

Go With the Flow

Education

University of Central Florida

Ph.D. in Computer Engineering [GPA: 4.00 / 4.00]

Jan 2021 - Present
Orlando, USA

Research	: Real-time Autonomous Systems
Advisor	: Dr. Zhishan Guo
Co-advisor	: Dr. Yaser P. Fallah
Roles & Achievements	:

Graduate Research Assistant.
Graduate Teaching Assistant: EECS 140 Introduction to Digital Logic Design.
Recipient of Graduate Dean's Fellowship.
Recipient of Hoglund Fellowship.

University of Central Florida

M.S. in Computer Engineering [GPA: 3.84 / 4.00]

August 2016 - August 2018
Orlando, USA

Research	: Wireless vehicular networks, Vehicle Safety, ADAS, Intelligent Transportation Systems
Thesis	: Real-time SIL Emulation Architecture for Cooperative Automated Vehicles Vehicles
Advisor	: Dr. Yaser P. Fallah

University of Mumbai

B. Eng. in Electronics Engineering [GPA: 3.90 / 4.00]

August 2010 - May 2014
Mumbai, India

Ranked 1st amongst 120 students in the Electronics dept.
Second year representative and Head of creative team at Annual college festival — Pegasus.

Experience

AI Research Engineer

NHK International Corporation

November 2018 - July 2020
Michigan, USA

Leading the development of Industrial automation system based on Deep Reinforcement Learning and Robotics

Graduate Research Assistant

CAVREL at UCF

Feb 2017 - August 2018
Orlando, USA

Advance Vehicle Emulator — Sponser: Ford Motor Company

The first of its kind Connected Vehicle Emulator (and Channel Simulator) system designed to tackle challenges in the testing of Safety Applications for Cooperative Vehicles. The main idea behind this project is to simulate a network of multiple virtual vehicles using a single Broadcaster (DSRC or V2X-LTE) that appear realistic to the vehicle under test. This not only provides easily configurable test scenarios but also provides flexibility to debug applications by repeating exactly same scenario.
Small-scale Connected Autonomous Vehicle — Sponser: CAVREL

Mentored a team of 5 undergraduate students to build a fleet of vision sensors equipped small-scale autonomousvehicles to navigate using advanced planning algorithms and thus provide a test-bed for V2X safety applications.
Vehicle Safety Communications Applications — Sponser: CAMP

The research goal is to verify the performance of connected cooperative vehicular safety system using Dedicated Short Range Communications (DSRC) communication over 5.9GHz dedicated wireless channel under both highly congested and uncongested traffic condition.

Machine Learning Intern

GIANT Health Events

May 2017 - June 2017
London, UK

Built highly accurate models based on SVM and Regression techniques to predict potential event attendees from the scrapped social network data and thus reduce human efforts and increase business.

Business Intelligence Developer

Tata Consultancy Services

September 2014 - July 2016
Mumbai, India

Migrated 95 high-complexity reports into production (Business Objects and Crystal reports to Microstrategy) within 3 months, along with a team of four members (Agile methodology)
Developed Interactive reports in Microstrategy to calculate and analyzefinancial tax for ABN-AMRO bank

Research & Academic Projects

Autonomous Vehicles

NSL | Udacity Nanodegree

Drivers Behavior Cloning

- Designed a CNN model to predict Steering wheel angles based on the Manual Driving Behavior
- Achieved a Validation loss of as low as 0.35%
- Autonomous Driving mode was tested on the Udacity's Unity based Self-Driving Simulator
- Driving behaviour involved Center-Lane driving, Sharp turns, Visual changes to road and surroundings

Advanced Tracffic Sign Classication

- Built and Fine-tuned a Convolutional Neural Network to classify traffic signs from the German Traffic Sign Dataset
- Preprocessed the training set images to generate Augmented images
- Attained an accuracy of 95% on Test Dataset and 97% on Validation Dataset
- Linked the model with a camera application to classify traffic signs in real-time

Vehicle Detection and Tracking

- Trained an SVM classifier to distinguish between car and non-car images with 98.56 % accuracy
- Streamed Sliding bounding boxes of different scales to detect presence of a vehicle
- Developed a heat-map of all positively detected vehicles to remove false positive based on a threshold

Real-Time Lane Detection

- Extracted Lane lines from multiple color spaces
- Fitted the extracted lane points into a polynomial
- Applied a sliding window recursively to find lane lines in every successive frame

Intelligent Agents

University of Central Florida

Autonomous Rescue Robot

- Built a 4-wheeled autonomous car for search and rescue operations in a disaster-affected area to identify victims
- Programmed ROS (Robot Operating System) nodes for gathering the odometry data along with the scans from Kinect sensor(to create Octotree and Occupancy maps) into Raspberry-Pi
- Implemented a particle filter for localization and a path planning algorithm for navigation to various goals using offline maps created during the training phase
- Two-way control - Teleoperation or Autonomous

Path Planning and Q-Learning in a grid world

- Implemented A* path planning algorithm with Manhattan and Euclidean distance choice along with an Interactive grid world GUI using python’s Tkinter library
- Designed a Reinforcement/Q-learning engine with both, deterministic and stochastic behavior in the same grid world

Robotics and Control

University of Mumbai

Surveillance Based on Tracking and Targeting

Cellphone Controlled RoboCar

Miscellaneous

A Concurrent Physics Engine

- Measured the performance of the different data structures in the context of a real-time physics engine
- Physics engine will move circles with random velocities around a screen. When a circle collides with the edge of the screen or another circle, it will bounce off and travel in another direction. The physics engine will divide time into equal slices called frames. Each frame will do the following in this order: Kinematics, Collision Detection, Collision Response.
- Implemented concurrent versions of the SAP list and hash grid. The implementations are tested in a physics simulation with moving objects where the data structure must detect objects that are near a point of interest.