Regulatory and Safety Standards for ML-Driven IoT Robots Operating in Hazardous Industrial Environments
TL;DR
A structured framework to ensure safety and compliance was designed, incorporating real-time risk assessment, ML-based threat detection, IoT security mechanisms, and pre-defined safety rules, and was well rated by experts on the completeness, practicability, applicability and regulatory fit.
Abstract
As the number of robots using machine-learning (ML) algorithms and connected to the Internet of Things (IoT) continues to grow in dangerous workplaces, the development and adoption of regulations and safety safeguards to ensure their use are not keeping up. Most standards on the industrial robot safety, functional safety, IoT cybersecurity and trustworthy AI stand alone, and there is no common and risk-based method to evaluate and ensure the safe use of autonomous robots learning to operate with human operators in risky environments. This study used the Design Science Research (DSR) method to design and test a risk-based regulatory and safety approach to ML-enabled IoT robots in hazardous industrial settings. The research first investigated the key safety hazards of autonomous robotic applications, such as communication failure, inaccurate ML predictions, sensor inaccuracies, unauthorised access, collision hazards, and unsafe decision making, and systematically analysed and mapped the safety requirements, regulation and technical standards relevant to industrial robots, IoT security, machine learning, functional safety, and autonomous systems to the identified hazards. In light of this analysis, a structured framework to ensure safety and compliance was designed, incorporating real-time risk assessment, ML-based threat detection, IoT security mechanisms, and pre-defined safety rules. Empirically observed results would not be reported or interpreted here, but are shown and interpreted as hypothetical, demonstrating how the accuracy of risk detection, safety compliance coverage, response time, reliability, reduction in unsafe robotic actions, and expert validation ratings would be reported if there were genuine simulated-scenario testing and expert review. The pattern showed that the proposed framework consistently attained high risk-detecting accuracy and safety compliance coverage in various simulated hazard situations and was well rated by experts on the completeness, practicability, applicability and regulatory fit.