Data-Driven Cellular Network Selector for Vehicle Teleoperations

TL;DR

This work presents an algorithm, called Active Network Selector (ANS), which uses a time series machine learning approach for solving the problem of video-based teleoperation, and compares it to a baseline non-learning algorithm, which is used today in commercial systems, and shows that ANS performs much better.

Abstract

Remote control of robotic systems, also known as teleoperation, is crucial for the development of autonomous vehicle (AV) technology. It allows a remote operator to view live video from AVs and, in some cases, to make real-time decisions. The effectiveness of video-based teleoperation systems is heavily influenced by the quality of the cellular network and, in particular, its packet loss rate and latency. To optimize these parameters, an AV can be connected to multiple cellular networks and determine in real time over which cellular network each video packet will be transmitted. We present an algorithm, called Active Network Selector (ANS), which uses a time series machine learning approach for solving this problem. We compare ANS to a baseline non-learning algorithm, which is used today in commercial systems, and show that ANS performs much better, with respect to both packet loss and packet latency.

Figures (13)

• Figure 1: The results of our communication test drives: Red spots indicate high packet loss in the Loss column and poor signal quality in the RSRP column; green spots indicate high velocity in the Speed column; the Handover column indicates the number of handover events that occurred in each location
• Figure 2: Latency comparison of three different cellular networks (Modems 1, 2, and 3) over a period of time. It is evident that no single network can offer latency shorter than 100ms during any of the considered time intervals
• Figure 3: The pipeline stages in the proposed ML framework
• Figure 4: A correlation matrix that represents the relationships between pairs of features in a dataset of communication drives; each value in this matrix represents the strength of the relationship between the two corresponding features;
• Figure 5: The prediction dataset of HandPredict for 80 test drives (first configuration)