Image Classification with Knowledge-Based Systems on the Edge for Real-Time Danger Avoidance in Robots

Abstract

Mobile robots are increasingly common in society and are increasingly being used for complex and high-stakes tasks such as search and rescue. The growing requirements for these robots demonstrate a need for systems which can review and react in real time to environmental hazards, which will allow robots to handle environments that are both dynamic and dangerous. We propose and test a system which allows mobile robots to reclassify environmental objects during operation in conjunction with an edge system. We train an image classification model with 99 percent accuracy and deploy it in conjunction with an edge server and JSON-based ruleset to allow robots to react to and avoid hazards.