Abstract
Floor-cleaning robots are becoming famous nowadays, and many robots are operating in public places autonomously as professional cleaners. Although there is a steep increase in the number of cleaning robots operating in public areas and continuous research development, these robots often face performance degradation in parameters including area coverage, safety, and long-term deployment due to the reason of inability in understanding the environmental context. There are so many context-aware perception algorithms that have been proposed for robotic systems that focus on various domains and applications. However, none of the existing algorithms targeted the context of cleaning or cleaning robots which provided a huge opportunity for research and development. This thesis presents a significant step in advancing knowledge toward a context-aware perception scheme that can aid the cleaning robot to understand the environment to plan its task accordingly for achieving maximum efficiency. The developed perception system uses various deep learning methods to learn information from vision, and acoustic sensors to understand the environmental context. As part of this study, two distinct robots were utilized which are beluga a differential drive cleaning robot, and Toyota HSR (human support robot) which will be deployed in an environment wherein these robots use the proposed context-aware perception method to perform effectively. The major contribution of this research work includes the development of A lightweight vision-based Deep Convolutions Neural Network (DCNN) framework to recognize the context on the top table by detecting the litter spillage and performing efficient cleaning. The developed system was evaluated in real controlled settings and achieved a 96% accuracy in terms of environment context detection and cleaning efficiency. This work also contributes to the development of a deep learning-based context-aware multi-level information fusion system for autonomous mobile cleaning robots to detect and avoid hazardous objects in an operational environment. The developed system demonstrated a higher confidence level and better detection under different levels of occlusion. This study also covers the development of a novel context avoidance framework based on a deep-learning method that can detect and classify a specific sound and localize the source from a robot’s frame to avoid that environment. The system was tested in various scenarios, the developed system accomplished a significantly higher success rate in detecting the unsafe context and avoiding it. The experimental results in all our trials show that the context-aware perception method could significantly improvise the overall performance of cleaning robots that includes area coverage, time is taken to clean, safe, and so on in any given context. This research work presents a comprehensive work on how the context-aware iii perception method could improvise the performance of a 2D dimensional cleaning robot which could be starting point towards achieving a learning-based-autonomous context-aware perception framework to improvise cleaning performance in a 3D space.