Optimising robotic operation speed with edge computing over 5G networks: Insights from selective harvesting robots

TL;DR

This paper investigates speeding up robotic strawberry harvesting by integrating edge computing over a private 5G network (E5SH). It systematically compares semantic segmentation models (Detectron2/Mask-RCNN and D2Go FBNet) on edge versus onboard hardware, and evaluates 5G versus WiFi and MQTT versus TCPROS in a field setting, quantifying latency, throughput, and energy use. The results show that edge-based inference with high-performance models delivers substantial speed-ups (up to 18.7x) over embedded processing, with 5G providing more stable, higher-throughput communication, and MQTT generally outperforming TCPROS. The study also analyzes energy and carbon implications, indicating that scalable edge deployment (serving 10–12 robots) can be more cost-effective and environmentally favorable than deploying individual onboard processors, paving the way for human-cost parity in commercial robotic cropping systems.

Abstract

Selective harvesting by autonomous robots will be a critical enabling technology for future farming. Increases in inflation and shortages of skilled labour are driving factors that can help encourage user acceptability of robotic harvesting. For example, robotic strawberry harvesting requires real-time high-precision fruit localisation, 3D mapping and path planning for 3-D cluster manipulation. Whilst industry and academia have developed multiple strawberry harvesting robots, none have yet achieved human-cost parity. Achieving this goal requires increased picking speed (perception, control and movement), accuracy and the development of low-cost robotic system designs. We propose the edge-server over 5G for Selective Harvesting (E5SH) system, which is an integration of high bandwidth and low latency Fifth Generation (5G) mobile network into a crop harvesting robotic platform, which we view as an enabler for future robotic harvesting systems. We also consider processing scale and speed in conjunction with system environmental and energy costs. A system architecture is presented and evaluated with support from quantitative results from a series of experiments that compare the performance of the system in response to different architecture choices, including image segmentation models, network infrastructure (5G vs WiFi) and messaging protocols such as Message Queuing Telemetry Transport (MQTT) and Transport Control Protocol Robot Operating System (TCPROS). Our results demonstrate that the E5SH system delivers step-change peak processing performance speedup of above 18-fold than a stand-alone embedded computing Nvidia Jetson Xavier NX (NJXN) system.

Figures (14)

• Figure 1: System process pipeline: An overview of the E5SH system process pipeline overview, processes in yellow are field-tested during E5SH experiments. The processes shown in red were not completed in field testing due to strawberries' out-of-season time.
• Figure 2: Detailed system architecture of the proposed setup : Illustration of the data flow between robot and edge-server over 5G and WiFi. The edge-server side performs 4-class classification. The computing device on the robot performs localisation, OctoMap generation and action planning. The robot performs the picking action, collects the images, and sends them to the edge-server.
• Figure 3: E5SH Image annotation: (a) Original image; (b); semantic annotation at the later season (September) of strawberry under direct/passive solar illumination; (c) depth image; (d) equivalent bounding box annotation.
• Figure 4: The basic network topology for communication from the robot to the edge-server for (a) 5G network and (b) WiFi network.
• Figure 5: The robot and polytunnel facility at the University of Lincoln: A Franka Emika arm equipped with the strawberry picking end-effector parsa2023autonomous sits on a Thorvald II. Cameras, communication devices, and laptops are also shown.