Answers are grounded in this paper and cite their sources.
The convergence of mobile robotics with telecommunications has driven the emergence of the Internet of Robotic Things (IoRT). However, deploying autonomous robots in unstructured environments faces a critical limitation: the unpredictable latency and high costs of the traditional Cloud Robotics model make it unfeasible in rural, agricultural, or remote industrial contexts with only low-speed network access. This paper presents a distributed Edge–Fog architecture for geo-referenced path tracking over low-cost LPWAN networks. The Fog layer, orchestrated with Node-RED, handles route planning (A*) and spatially compresses trajectories (String Pulling) to make transmission over the limited LoRa channel feasible. The Edge layer, based on an STM32 microcontroller running FreeRTOS, locally manages GPS/IMU heading estimation, heading tracking through a proportional–derivative (PD) controller with Cross-Track Error (XTE) correction, and reactive obstacle avoidance via a Sugeno-type Fuzzy Inference System (FIS). Waypoint delivery relies on a Stop-and-Wait ARQ protocol with automatic retries and duplicate tolerance. Four outdoor missions at a campus site, with one run per scenario and gateway–robot distances of 54–74 m, show that: (i) String Pulling reduced the number of route points by 93.3–96.4%, so that each route was loaded with one or two waypoint frames (4–5 acknowledged exchanges); (ii) all waypoints were stored correctly in the robot in the four field transfers (route-level delivery ratio of 100% after ARQ, with two retransmission timeouts and no dropped frames), and the over-the-air load took 1.91–3.03 s, while single-run controlled fault-injection bench tests showed automatic frame retransmission, duplicate-payload suppression on the STM32, and detection of an empty route; (iii) three field missions completed autonomously at the final target, and one (M3) was manually halted right before the arrival radius of the final waypoint due to an unplanned obstruction near the endpoint, with a mean XTE of 0.98–1.28 m over the four missions (the M3 statistics cover the telemetry up to the halt), consistent with the accuracy of the low-cost GPS receiver; (iv) in a single additional field trial with an intentional 87.52 s LoRa blackout (63.8% of the 137.27 s mission), the robot continued navigating autonomously, resumed telemetry reporting upon reconnection, and completed the mission autonomously (mean XTE of 1.01 m); and (v) static and dynamic obstacles were avoided in separate qualitative tests. These results are limited to the tested conditions: the individual contribution of each control component and the quantitative advantage of the FIS over a binary reactive controller were not evaluated experimentally.
No takes yet. Share an insight, caveat, or question.
Maldonado-García et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: