IoT-Based Remote Monitoring and Control of Robots for Hazardous Environments Using Machine Learning

Le résumé fourni par la source

Robots that are used in dangerous places, including chemical plants, mines and disaster-stricken areas, must work reliably and keep the human operator away from the danger. This study proposes and tests a system based on Internet of Things (IoT) that allows remote monitoring and control of robots in hazardous environments, complemented by a machine learning (ML) algorithm to automatically classify the hazard. The proposed system combined the environmental sensors (battery level, motor current, and vibration) with the environmental sensors (temperature, humidity, gas concentration, and smoke) and an edge-computing/microcontroller unit with an IoT communication module and a remote monitoring-and-control interface installed on a mobile robotic platform. A supervised machine learning (ML) model was trained, to classify the sensed environment into three safety classes: Safe, Caution and Hazardous with 4 candidate algorithms (Random Forest, Support Vector Machine (SVM), Decision Tree, Artificial Neural Network (ANN)) evaluated on accuracy, precision, recall, F1 score and inference time with 3000 simulated sensor data. The SVM classifier had the highest overall accuracy (90.0%) and F1-score (.890) compared to the Random Forest (89.5% accuracy) and ANN (88.9% accuracy) classifiers, with the Decision Tree also providing the fastest inference time (0.09 ms) at a relatively small cost in accuracy (86.5%). Overall, experimental tests conducted at the system level revealed that the end-to-end communication latency was highly meaningful with hazard intensity (p < 0.001; F = 335.1), ranging from a mean of 119 ms for the lowest hazard intensity to 210 ms for the highest hazard intensity, and that command-execution success rate and real-time monitoring accuracy decreased moderately as hazard intensity increased (p < 0.001; F = 205.5 and 72.9, respectively), but remained above 83% and 92% for the highest hazard intensity, respectively. These results have shown that the combination of IoT communication, edge sensing, and the classification of hazards by machine learning can realize low-latency and reliable monitoring and control of robots in hazardous environments and also show that there is a measurable performance trade-off when increasing the intensity of the environment hazards.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.