Masters Thesis

Destination Selection Using Deep Learning for an Autonomous Wheelchair Navigation in an Unmapped Indoor Environment

Recent developments in robot automation have fostered the development of many assistive devices to improve the quality of life for individuals with disabilities. Notable among these devices are autonomous wheelchairs, which are capable of navigating to given destinations while avoiding obstacles. However, the method of destination selection and navigation in unmapped indoor environments remains a challenge for these autonomous wheelchairs. in this work, a novel approach to selecting a destination for an autonomous wheelchair in an unmapped indoor environment using Deep Learning System for object detection, camera, ranging LIDAR is presented. Object Detection Process not only recognizes and classifies object of interest in an image but also localizes each object by along with a bounding box and desired class name around it. the model for object detection was developed on TensorFlow Framework using three classes of dataset ie Forever21, JCPenney and Macy’s. the proposed System scans the environment (prototyped mall) at startup and compiles a list of possible destinations which are fed to object detection system for Predictive Comparison of the Stores. Once the stores are detected, Lidar examines the distance and the wheelchair is successfully able to navigate to the destination along with obstacle avoidance. the system successfully navigated to the destination in100% of the trials for close-range destinations and 90% of the trials for mid-range and long-range destination.


In Collection: