Graduate Research Assistnat

Shivang Patel

ROBOTICS ENGINEER

+1-240-***-**** adckp0@r.postjobfree.com shivangapatel.com shivaang12 shivaang14 @Shivang_engg Education

University of Maryland College Park, MD 20740

M.ENG. IN ROBOTICS 2017 - 2019

G. H. Patel College of Engineering Vallabh Vidhanagar, India B.ENG. IN MECHATRONICS ENGINEERING 2013 - 2017

Technical Skills

Proficient in C++, Python, ROS, ROS 2, Gazebo, Ignition, openCV, Tensorflow, Keras, PyTorch, GIT, Docker and Linux. Experience

University of Maryland College Park, MD 20740

GRADUATE RESEARCH ASSISTANT, AEROSPACE ENGINEERING DEPARTMENT April 2018 - Present

• Involved in the development of the DARPA’s Offensive Swarm Enabled Tactics (OFFSET) project (sprint 1) which utilizes more than 100 small Un- manned Aircraft Systems (UASs) to accomplish diverse missions in a complex urban environment.

• Researched two global planners which can optimize the area coverage in any complex urban environment and developed a compatible obstacle avoidance method for one of them which was lacking previously.

• Global Planners were developed in C++ as ROS packages and maintained using GIT with Doxygen style documentation. University of Maryland College Park, MD 20740

GRADUATE RESEARCH ASSISTANT, MECHANICAL ENGINEERING DEPARTMENT Oct 2017 - April 2018

• Designed and developed an acoustic sensor using Metamaterials which amplifies wave pressure, makes detection of low wavelength possible.

• Variation of designs were prepared using CAD software SOLIDWORKS; Performed various levels of testing on the sensor designs for validation.

• Used LAUNCHXL-F28379D prototyping board for the sensor system integration with the functional mobile robot for obstacle avoidance. Projects

Self Driving Toy Car using Reinforcement Learning Technique (git.io/Jv4D0)

• Trained an RC Toy car using Deep Q Learning with python 3 and tensorflow library, to make it learn to drive itself in a lane.

• Project contains two systems: toy car - modified to be controlled by Raspberrypi with a front camera; and a mainframe computer for training.

• Used ROS for system integration; RaspberryPi to send camera feed to mainframe PC which trains the Deep Q Learning network and in return would send the control commands to the toy car; Reward and Penalty were given manually in each training iteration. Smart AGV (github.com/shivaang12/smart_agv)

• Leveraged ROS Navigation stack to augment the Relaxed Astar algorithm as a global planner plugin for the autonomous traversal of an AGV

• Followed Software Development Techniques like Solo Iterative Process, maintaining product backlog, iteration backlog andworklogandgenerated Doxygen style documentation.

• Used Git as a version control system, designed Unit tests, employed Continuous Integration through travis.ci, and used tools like cpplint (check styling errors), cppcheck (static code checking) and valgrind to check memory leaks. Evaluation of Adversarial Attacks and Defenses for DNN (git.io/Jv4D1)

• Selected popular adversarial attacks, such as FGSM, JSMA, DeepFool, NewtonFool, Carlini and Wagner attack and Basic Iterative Method, and de- fenses, such as GDA, Label Smoothing, Feature Squeezing and Spatial Smoothing, and evaluated their performance on various networks, such as two-layer, four-layer, seven-layer, ten-layer, two-layer convolutional net and four-layer convolutional net.

• Tested two hypotheses - no defense method will allow classifiers to be robust against all attacks, and adversarial efficiency will decrease as the network gets deeper.

• Used MNIST dataset to test these hypotheses. Synthesize results into conference paper and presented to mock conference. Traffic sign detection using feature detection methods and Machine learning routines

• First stage includes detection, which was achieved using image processing techniques such as RGB to HSV conversion, morphing, masking, contour detection and so forth.

• Second stage is recognition which consists of pre-processing, training a Support Vector Machines (SVM) classifier after feature extraction using His- togram of Oriented Gradients (HOG).

Open Source Project

Navigation 2 (github.com/ros-planning/navigation2))

• Actively contributing to Navigation2 ROS2 package on Github and participating in Navigation2 Work Group weekly meetings.

• My contribution includes improving documentation, Porting changes from ROS 1 libraries, Bug improvements and code review. APRIL 1, 2020 SHIVANG A. PATEL · RÉSUMÉ 1

Contact this candidate