| MS TSA Meeting

Name of the Speaker: Mr. Soumyajit Chakraborty (EE21S112)
Guide: Prof. Ramkrishna Pasumarthy
Venue: ESB-234 (Malaviya Hall)
Online meeting link: https://meet.google.com/uxp-uuzb-enh
Date/Time: 5th June 2024 (Wednesday), 11:00 AM
Title: Reinforcement Learning-Driven End-to-end Navigation Systems for Autonomous Vehicles and Mobile Robots

Abstract :

Autonomous navigation systems are crucial for autonomous vehicles and mobile robots, enabling them to make real-time decisions, plan paths adaptively, and navigate through cluttered and dynamic environments without human intervention. We present an end-to-end autonomous driving architecture designed for navigating heterogeneous highway traffic using reinforcement learning (RL) algorithm. The architecture facilitates the development of an RL agent capable of making driving decisions directly from sensor data. We evaluate the driving performance of the autonomous vehicle (AV) using two types of sensor inputs: top-view images of the AV and its surrounding vehicles, and data on the relative positions and velocities of surrounding vehicles with respect to the AV and compare their efficacy under varying traffic densities. Furthermore, we extend this end-to-end navigation system to mobile robot navigation within a closed indoor environment, where the robot must navigate from its current position to a desired location without prior access to a map of the environment or any prior knowledge of obstacles. We implemented our algorithm on the Turtlebot hardware robot platform for validation and compared its performance against sampling-based motion planning algorithms that require a previously mapped environment. Our algorithm demonstrated reduced travel distance compared to traditional methods, assuming other parameters remained constant.