Experimental Characterization of ROS 2 Communication over Wireless Networks for Robot Control

This thesis presents an experimental characterization of ROS 2 communication latency over a real wireless link, using a laptop and a Raspberry Pi 4 connected over Wi-Fi 4 (IEEE 802.11n, 2.4 GHz). The experiments examine the effects of message payload size, hardware role allocation, network-usage conditions, and subscriber count on the latency distribution. The results show that ROS 2 latency over Wi-Fi is right-skewed and can exhibit an upper tail, especially under higher load. Increasing the number of subscribers from one to twenty shifts the latency distribution toward higher values and increases variability. To characterize the empirical latency behavior, five candidate probability models were fitted to the latency data: Normal, Log-Normal, Gamma, Weibull, and a two-component Gaussian Mixture Model (GMM). Model selection using AIC and BIC consistently identified the two-component GMM as the preferred model across the tested subscriber-count datasets. The GMM captures both the dominant low-latency region and the higher-latency tail more effectively than the single-distribution alternatives. The GMM parameters were regressed against subscriber count using quadratic polynomial regression, yielding a compact latency model within the calibrated range of one to twenty subscribers. The results show that average latency alone is not sufficient for evaluating wireless ROS 2 communication; percentile-based and distributional metrics are necessary for time-sensitive robotic applications.