Engineer Engineering

YASHWANTH SURAPANENI

513-***-**** ********@****.**.*** https://www.linkedin.com/in/surapayh

EDUCATION

Masters: Robotics & Intelligent Autonomous Systems GPA: 3.875/4.00 Aug 2018 - Dec 2020 University of Cincinnati, Cincinnati OH

Bachelors: Mechanical Engineering GPA: 7.21/10.0 July 2013 - May 2017 Visvesvaraya National Institute of Technology, Nagpur, India RELEVANT COURSEWORK

Decision Engineering, Robot Control & Design, Bio-Inspired Robotics, Intelligent Autonomous Mobile Robots. TECHNICAL SKILLS

Python (TensorFlow, PyTorch, OpenCV, SciPy, NumPy, Matplotlib), MATLAB, C++, Linux, ROS, GitHub, Gazebo, Rviz. WORK EXPERIENCE

Object Detection/Computer Vision Intern, Medha Servo Drives Private Limited, Hyderabad, India Mar 2018 – July 2018

• Worked on the Robotics team and involved in the Autonomous Driving - Object Detection project using CNNs.

• Real time car detection data from the roads was collected with over 80 different classes. Implemented pre-trained YOLOv2 model to detect these 80 classes.

• Data was preprocessed and the CNN (Convolutional Neural Network) was trained using TensorFlow.

• Output was filtered to achieve high accuracy in Object detection and Classification Graduate Engineer Trainee, Varroc Engineering Limited, Aurangabad, India July 2017 - Feb 2018

• Improved seat runout in engine valves after hardening through continuous mesh belt furnace reducing the rework.

• Modified the valve handling system resulting in less dents and damages of the finished valves.

• Optimized the friction welding process leading to increase in weld strength in the exhaust valves. CAPSTONE PROJECT

Indoor Application Mobile Robot using NVIDIA Jetson Nano (Hardware) Obstacle Avoidance

• Addressed the problem of obstacle avoidance in mobile robots for indoor applications. Raspberry Pi V2 camera was used to collect data to train the neural network. Pre-trained Alex Net was further trained using PyTorch to make the turn decision if there was an obstacle present in the workspace of the robot. GPU-Accelerated Object Tracking

• Pre-trained ssd_mobilenet_v2_coco is sourced form TensorFlow Object detection API which is trained over 90 different classes. The network is optimized using NVIDIA TensorRT for the real-time application on Jetson Nano.

• The ORB features of the detected objects along with their descriptors are extracted and tracked over every frame to describe their motion.

ACADEMIC PROJECTS

EKF-SLAM (Simultaneous Localization and Mapping) using April Tags and RGB-D Camera in ROS-Gazebo

• Turtlebot3 with noisy input and noisy sensor information is passed through a set of way points with minimum error in an obstacle filled unknown environment with April tags as landmarks of both known and unknown correspondence. Digital Image Classification using Transfer Learning

• Trained an autoencoder on MNIST data set to obtain meaningful features of handwritten digits using momentum, sparseness, and weight decay. The features are transferred to a feed forward network to improve classification. Self-Driving Car in GTA V game engine

• Training data was collected by moving the car on the road in the game engine. Input is the live window screen and output is the command to move. Pre-trained Alex Net is further trained using TensorFlow to achieve self-driving. Hand Sign Classification using ResNets

• A very deep neural network of 50 layers is trained in TensorFlow without the problem of vanishing gradients to achieve classification of hand signs with a very high accuracy. using skip connections and average pooling. Art Generation using Neural Style Transfer in CNNs

• A new image is generated by combining a content image whose contents are to be retained in the generated image and a style image whose art style must be retained in the generated image using a pre-trained VGG-19 model in TensorFlow. ORB- Feature tracking using Raspberry Pi V2 camera

• ORB features were extracted along with their descriptors for the images collected in different lighting conditions and orientations. The corresponding features were tracked using Brute-Force matcher based on Hamming distance. WORK AUTHORIZATION

Eligible to work in U.S. with Optional Practical Training.

Contact this candidate