Autonomous Systems Project

• Motor Speed Control is done using Pulse Width Modulation.

• Hardware used: IG42 Geared Motor (24V 122 RPM) with quadrature encoder, Arduino Due, Cytron MD10 driver.

• Used Arduino Matlab interface to obtain step response of the Motor to PWM step input.

• Based on Step Response, Transfer function was built and the gains were developed.

• Part of the Autonomous Ground Robot Project

• Used a Faster RCNN model trained on the KITTI dataset to detect Cars and Pedestrians.

• The detector was trained using transfer learning using Tensorflow Object Detection API

• Runs at 12 FPS

• This is a Tracking using Detection Algorithm that uses Hungarian Algorithm for Data Association

• The Detected results are tracked using IOU and Euclidean distance as the matching metric.

• Runs at the same 12 FPS when running with the detector.

• Currently working on correlation trackers and Kalman Filter based Trackers.

• Udacity course project.

• Start with camera calibration to obtain the intrinsic and extrinsic parameters using checkerboard technique.

• After removing distortion, converted image to HLS and thresholded using Sobel Algorithm.

• From the bottom pixel lane point, we use sliding window search to capture other points on the lane and use polyfit to fit a three degree polynomial.

• Results are shown beside.

Simple Hough Transform Based Lane Detection system that extracts keypoints using Canny edge detection and uses Open CV to detect lane lines using Hough Space.

Developed Stanly control for Lateral control and a simple PID controller for Longitudinal actuator control based on a set of waypoints generated using pre-programmed motion planner. This is the course project for Self Driving Cars introduction in Coursera.

• Created a Finite State Machine Behavioral Planner for a simple stop sign scenario

• Final Project of Self Driving Car Specialization

Using the Unity player and Keras to train neural network to train driving of a car by manually driving the car and feeding steering angle lane position as training data.

Trained a Conv Net using Keras to classify Traffic Signs using the GTSRB Dataset

Developed a 11DOF vertical ride model for simulation of various control strategies of semi active suspension.