Machine Learning Based Adaptive Link Quality Prediction for Robot Network in Dynamic Environment

The usage of wireless networks to teleoperate a robot in operational scenarios or safety-critical applications has increased significantly in recent years. In these environments, it is important to maintain a steady connection between the control center and the robot. One of the ways to ensure this steady connection is to estimate the Link Quality (LQ) between the robot and control center. Estimating the LQ can help in alerting an operator about a potential failure in a communication link. However, it is costly in terms of network resources to transmit control packets that will track or monitor the LQ over the network (i.e., it increases the network traffic). It becomes a more difficult task to estimate LQ when mobility of the robot is a constraint and the environment in which the robot is functioning is dynamic. In this work, we propose an offline machine learning based model that detects the operational environment of a robot using both radio and network parameters. Based on this environment, LQ is predicted. Our results show that we can use received signal strength, signal quality, expected transmission count and communication distance to predict LQ and characterize the operational environment of a robot. We achieved an F1-score of 81% on the test data using our model to classify the operational environment. The mean absolute error of 0.09 was obtained for predicting Throughput Potential Ratio (TPR).