CCS335 _ Neural Networks and Deep Learning Laboratory_Lab Complete Record
Kaustav_Chakraborty_resume
1. EXPERIENCE
Innovation Chair, Human Powered Vehicle Student Team, VIT University, Vellore, India May 2016 – Mar. 2018
· Lead a team of five in the implementation of innovative safety features based on proximity sensors with embedded systems.
· Managed two team-departments in CAD modelling, FEA, and coding in C.
Plant Automation and Calibration Intern., HPCL, India Dec. 2016 – Feb. 2017
· Calibrated and maintained large datasets for the daily operation of centrifugal pumps, Variable frequency drives, flowmeters.
· Applied regression techniques to estimate operation costs and brought the margin of error to 0.75% to show 35% savings in operations .
EDUCATION
· University of Michigan, GPA:3.95/4.0 Ann Arbor, MI
Master of Science in Engineering, Robotics Sept. 2018 - Apr. 2020
Coursework: Self Driving Cars, Computer Vision, Computational Data Science, Machine Learning,
· VIT University, CGPA: 9.35/10 Vellore, India
Bachelor of Technology in Mechanical Engineering May 2018
Coursework: Industrial Automation controllers, Mechatronics, Data Structure and Algorithms,
: vatsuak@umich.edu : github.com/vatsuak : (734)-8813359 : kaustavchak
KAUSTAV CHAKRABORTY
RESEARCH
Graduate Student Research Assistant, Bipedal Robot Laboratory, Ann Arbor Michigan May 2019
Designed Deep Neural Net for LiDAR based Detection of Visual Fiducial scans with an accuray of 99.82% .
(1% improvement over the State of Art) in Tensorflow with CUDA architecture under Prof. Jessie Grizzle.
Handled over 80Gb of real world dataset collected by Cassie robot’s Vision Suite to produce online software.
Read Lidar data from Velodyne Ultra Puck using Velodyne ROS Package and stored raw returns as ROSBAG files.
Applied SVD, homogeneous transforms for dimentionality reduction and data projections for feature engineering.
Employed LSTM and PointNets to build novel convolutional RNN network.
Graduate Researcher, Biologically Inspired Robotics and Dynamical Systems Lab, Ann Arbor, MI. Nov. 2018
· Designed Self-righting shell for Terrestial Hexapod -“Big Ant” under Prof. Shai Revzen.
· Optimized Spline Designs in the form of 2nd order differential equations,parameterized with 6 variables using Matlab.
· Modelled 3D sturucture of shell inspried from insects using Solidworks and Audodesk Inventor .
CAREER OBJECTIVE
Goal oriented robotics engineer, looking for full-time positions in areas of perception in autonomous vehicles, computer vision, artificial
intelligence and robotics. Possesses a high aptitude for optimal design of systems.
PROJECTS
· Applied SLAM to mobile robotics using techniques like Monte Carlo Methods, Baeyes Localization, Kalaman Filters and its
derivatives.
· Designed JORB-SLAM, a multi-agent slam system that employs a centrlized optimization technique by running ORB-SLAM in one or
more agents, additionally using AprilTags based visual localization to provide infrormation and add additional linkages/keyframes to
the factor graph so developed, and tested the results on KITTI dataset as well as self-collected dataset.
· Implemented essential image techniques like feature detection, Visual bag of words, Panoramic stitching , Structure from motion and
camera calibration in Python using openCV, open3D.
· Implemented and trained a U-Net inspired model using Pytorch to segment Facade dataset achieveing 70% accuracy
· Employed transfer learining using RCNN, and VGG models for recognizing 5 different car classes in a data set of size 10 Gb sourced
from GTA 5 game achieving a prediction accuracy of 72%.
· Built and controlled a 4-DoF robotic arm coupled with Kinect depth-camera, forward and inverse kinematics, trajectory planning and
motion smoothing. Designed a rack-and-pinion based gripper.
· Employed methods of particle filtering using incoming sensor information from 2D RP-LiDAR to build a SLAM system for
autonomous wheeled robot. Implemented occupancy grid mapping and A* path planning.
· Designed a PID controlled cascading control loop to balance a bi-wheeled balancing robot, using wheel encoders and gyroscope for
implementing a unique gyrodometry based dead-reckoning system.
SKILLS
Languages : Python, C/C++, Julia, MatLab, JAVA, .
Frameworks: TensorFlow, Keras, OpenAI gym, Pytorch, OpenCV, Open3D, Scikit-Image, Scikit-Learn.
Others : Git/Version Control, SolidWorks, Autodesk Fusion, ANSYS, ADAMS, JSON, Allen Bradley PLC controller
Platforms : ROS, Linux-Ubuntu, Mac OS, Windows.