An Efficient Ground-aerial Transportation System for Pest Control Enabled by AI-based Autonomous Nano-UAVs

TL;DR

This work tackles efficient pest detection and treatment in crops by coupling a fleet of ultra-low-power nano-UAVs with a slow ground tractor. It deploys an edge-optimized CNN (SSDLite-MobileNetV3) on a GAP9 SoC to detect pests in real time at around $6.8$ frames per second while consuming about $33\,\mathrm{mW}$, achieving a mean average precision of $mAP=0.79$ on a 3-class insect dataset. The system uses a two-level routing strategy based on A*: a global planner for area-wide exploration and a local onboard planner (up to 50 Hz) for obstacle avoidance on a $4\times4\mathrm{m}$ occupancy map, with Webots simulations validating the approach across multiple environments. Results show the ground-aerial transportation system can cut exploration and treatment time by up to $\sim 20$ hours in a $200\times200\mathrm{m}$ vineyard and demonstrate a practical path toward autonomous precision agriculture with reduced pesticide usage and environmental impact.

Abstract

Efficient crop production requires early detection of pest outbreaks and timely treatments; we consider a solution based on a fleet of multiple autonomous miniaturized unmanned aerial vehicles (nano-UAVs) to visually detect pests and a single slower heavy vehicle that visits the detected outbreaks to deliver treatments. To cope with the extreme limitations aboard nano-UAVs, e.g., low-resolution sensors and sub-100 mW computational power budget, we design, fine-tune, and optimize a tiny image-based convolutional neural network (CNN) for pest detection. Despite the small size of our CNN (i.e., 0.58 GOps/inference), on our dataset, it scores a mean average precision (mAP) of 0.79 in detecting harmful bugs, i.e., 14% lower mAP but 32x fewer operations than the best-performing CNN in the literature. Our CNN runs in real-time at 6.8 frame/s, requiring 33 mW on a GWT GAP9 System-on-Chip aboard a Crazyflie nano-UAV. Then, to cope with in-field unexpected obstacles, we leverage a global+local path planner based on the A* algorithm. The global path planner determines the best route for the nano-UAV to sweep the entire area, while the local one runs up to 50 Hz aboard our nano-UAV and prevents collision by adjusting the short-distance path. Finally, we demonstrate with in-simulator experiments that once a 25 nano-UAVs fleet has combed a 200x200 m vineyard, collected information can be used to plan the best path for the tractor, visiting all and only required hotspots. In this scenario, our efficient transportation system, compared to a traditional single-ground vehicle performing both inspection and treatment, can save up to 20 h working time.

Figures (13)

• Figure 1: Picture of our nano-UAV prototype (A) and the GAP9Shield block diagram (B).
• Figure 2: Power breakdown of our robotic platform.
• Figure 3: The maximum detectable object speed depends on both the object dimension ($O_\text{side}$) and the distance from the sensor. $O_\text{side}$ is defined as the projection of the object on the sensor frame.
• Figure 4: SSDLite with MobileNetV3 backbone architecture.
• Figure 5: Samples of the three classes in our dataset 9601235.