AGRICAM: A Track-Mounted Crop Pollination Monitoring Robot
Authors: Malika Nisal Ratnayake, Adel N. Toosi, James Cook, Romina Rader, Alan Dorin
Organizations: Department of Data Science and Artificial Intelligence, Monash University, Wellington Road, Clayton, VIC, 3800, Australia · School of Computing and Information Systems, The University of Melbourne, Parkville, Melbourne, VIC, 3052, Australia · Hawkesbury Institute for the Environment, Western Sydney University, Locked Bag 1797, Penrith, NSW, 2751, Australia · School of Environmental and Rural Science, University of New England, Armidale, NSW, 2351, Australia
Insect pollination is critical for global food production, yet monitoring pollinators at commercial farm scale remains a challenge. Recent advances in computer vision and deep learning have enabled detailed analysis of pollinator behaviour, but monitoring must trade-off detail against spatial coverage and human or technological resources. This paper presents the Automated Guided Robot for Insect and Crop Activity Monitoring (AGRICAM), a purpose-built robotic system designed to meet the requirements of large-scale pollination monitoring in protected cropping systems. AGRICAM operates autonomously on low-cost, easily installed track for movement along crop rows, without disrupting farm operations or insect behaviour. The platform integrates two RGB cameras, microclimate sensors, GPS and RFID modules, motion sensors, and 4G cellular network connectivity for data transmission. A web interface enables remote device configuration and scheduling. The system autonomously captures video and image data of insects' locations and local environmental conditions. These are transferred to the cloud and analysed using computer vision models to quantify pollinator visitation and spatio-temporal activity variation. We deployed the system on a commercial blueberry farm to demonstrate and test its capability. It successfully mapped insect pollination patterns across 80 m long industrial polytunnels over 30 hours. This data enabled spatial analyses of insect activity we used to confirm a uniform pollinator distribution within polytunnels, as desired by the farm management team. The data also highlighted variation of insect activity associated with time of day and microclimate. AGRICAM therefore has been shown to be a scalable, automated crop pollination monitor that can support data-driven decisions to enhance pollination management, thereby improving crop productivity and food security.
Figures & tables
Figure 1: Overview of the AGRICAM processing pipeline. Video, location, and microclimate data are collected in the field using AGRICAM, transferred via cloud (or USB storage if desired), and processed by AI-facilitated workflows to automatically detect and track insect pollinators. The data are subsequently analysed and visualisations of spatiotemporal patterns of microclimate and insect behaviour are prepared.
Figure 2: AGRICAM system components . (a) System installation overview, (b) components of the AGRICAM robot, and (c) deployed system in a commercial blueberry farm. In (a) and (c), the AGRICAM unit is mounted on a PVC track along the leg rows of a protected cropping structure (polytunnel). Each monitoring station is identified by an RFID tag and positioned at structurally supported locations to minimise stress on the PVC track. In (b), task-specific components of the AGRICAM system are shown in different colours.
Figure 3: AGRICAM web user interface. (a) System status dashboard, (b) monitoring path map, (c) control and diagnostics panel, and (d) configuration menu.
Figure 4: AGRICAM field test site detail. (a) A map of the study site at Costa Berries Exchange, New South Wales, Australia. The location of managed beehives is indicated by a yellow pentagon. The crop rows where AGRICAM was deployed are labelled Row A and Row B, and the length of each monitored tunnel is indicated. Arrows show the direction the AGRICAM camera faces during row traversal. (b) Schematic representation of the tunnel configuration and blueberry row numbering, shown as viewed from the polytunnel end marked 0 m in (a). (c) Example image illustrating the field of view captured by AGRICAM.
Figure 5: Spatial and temporal distribution of video recordings during field deployment. Bubble plots show the number of videos recorded (colour scale 0–6) by distance from the polytunnel end (meters, x-axis) and by 2-hour time bin (09:00–, 11:00–, 13:00–, 15:00–16:59; y-axis) for two crop rows (Row A - green, Row B - blue) across two days. Panel titles report the total number of observations ( N ) for each day and row.
Row
Date
No. of
videos
Insect detection
Flower visit detection
Human obs.
TP
FP
FN
Human obs.
Software obs.
A
Day 1
3
6
5
2
1
7
3
A
Day 2
3
10
10
1
0
2
2
B
Day 1
3
13
10
1
3
8
8
B
Day 2
3
10
8
0
2
3
8
Table 1: Validation of automated insect and flower visit detection using the Polytrack software. “Row” and “Date” indicate the video samples analysed, and “No. of videos” the number evaluated. Under insect detection, “Human obs.” represents manually counted honeybees, while “TP”, “FP”, and “FN” denote true positives, false positives (including duplicate tracks of a single insect), and false negatives, respectively. Under flower visit detection, “Human obs.” indicates manually recorded insect–flower interactions and “Software obs.” those identified by the software.
Figure 6: Spatiotemporal variation in insect activity across two blueberry rows. Mean number of insect tracks (points ± 95% CI) recorded along the length of Row A (green, left column) and Row B (blue, right column) during four two-hour time blocks (09:00–17:00) across two sampling days. X-axis = distance from polytunnel end (m), y-axis = number of tracks. Crosses (x) denote raw values. The time period, and total number of tracks per row ( T ), appear above each panel.
Figure 7: Spatiotemporal variation in flower visits across two blueberry rows. Mean number of flower visits (points ± 95% CI) recorded along the length of Row A (green, left column) and Row B (blue, right column) during four two-hour time blocks (09:00–17:00) across two sampling days. X-axis = distance from the polytunnel end (m), y-axis = number of flower visits. Crosses (x) denote raw values. The time period, and total number of flower visits per row ( FV ), appear above each panel.
Figure 8: Distribution of insect tracks in relation to time of day and microclimate across two sampling days. Number of insect tracks recorded in Row A (green) and Row B (blue) plotted against time of day (hours), temperature ( ∘ C), and relative humidity (%) for Day 1 (top row) and Day 2 (bottom row). Each point represents the number of bee tracks observed during an individual sampling interval. Grey histograms show the total frequency distribution of non-zero insect track observations across each environmental variable. Red dashed vertical lines indicate the minimum and maximum values recorded for each variable during the sampling period. The left y-axis shows the number of insect tracks per observation, and right y-axis (histograms) shows the total number of non-zero insect track observations.
Agricultural robotics develops autonomous machines that sense, navigate, and perform tasks such as monitoring, weeding, spraying, and harvesting crops.
The decline of natural pollinators has created a major challenge for crop production in controlled indoor agriculture, particularly in vertical farming environments where natural insect pollination is absent. This motivates the development of robotic systems capable of performing precise flower targeting tasks while minimizing physical interference with delicate floral structures. This paper presents an aerial manipulator platform for perception driven flower detection, localization, and approach in vertical farming environments. The proposed system integrates onboard RGBD based perception, model predictive path integral (MPPI) based unmanned aerial vehicle (UAV) control on a PX4 platform, and a lightweight 2DoF manipulator for precise end effector positioning. The platform is evaluated in both MuJoCo simulation and UAV lab experiments using a flower targeting testbed. The experimental results demonstrate stable UAV flight, reliable flower localization, and centimeter level end effector positioning accuracy. In simulation, the proposed controller achieves consistent trajectory convergence and accurate target alignment. In the real world UAV lab environment, the integrated perception control manipulation framework enables stable flower targeted positioning and end effector alignment under constrained aerial operation. These results validate the proposed aerial manipulator as a robust robotic carrier and positioning framework for future contactless pollination systems. While the current study focuses on perception guided targeting and positioning, the developed platform provides a practical foundation for integrating advanced contactless end effectors, including acoustic based pollen manipulation modules, in future work.
Chenzhe Jin, Zhuohang Wu, Yifan Cai +4
Department of Computer Science University College London London, United Kingdom · Department of Artificial Intelligence University of Malaya Kuala Lumpur, Malaysia
Crop monitoring is essential for precision agriculture, but current systems lack high-level reasoning. We introduce a novel, modular framework that uses a Vision Language Model (VLM) to guide robotic task planning by actively querying heterogeneous data sources, including enriched RGB camera feeds and 2D semantic occupancy maps, interleaved with robotic action primitives. We contribute a comprehensive benchmark for short- and long-horizon crop monitoring tasks in monoculture and polyculture environments. Our results show that while zero-shot VLMs perform robustly for short-horizon tasks (achieving 87% success, comparable to human experts), success drops significantly to under 10% for complex long-horizon, multi-target tasks. Despite this decline, task completion rates remain above 76% under noiseless conditions. Critically, the system degrades when relying on noisy semantic maps, demonstrating a key limitation in current VLM context grounding for sustained robotic operations. This work offers a deployable framework and critical insights into VLM capabilities and shortcomings for complex agricultural robotics.
Jose Cuaran, Kendall Koe, Aditya Potnis +2
The Siebel School of Computing and Data Science, University of Illinois, Urbana-Champaign · National Center for Supercomputing Applications at University of Illinois, Urbana-Champaign · The Department of Agricultural and Biological Engineering, University of Illinois, Urbana-Champaign
The scalability of organic agriculture is partially limited by the labor costs associated with monitoring for pests. While drones and rovers are well-suited for agricultural monitoring from above or next to plants, many pests live on the underside of leaves or on plant stems, making them detectable only after they have caused significant damage. To enable early pest detection we present STEMbot, a miniature climbing robot system designed for autonomous navigation under plant canopies. Unlike existing climbing platforms that lack on-board perception or are restricted to unbranched vertical trunks, STEMbot integrates a fully geometric PIN-SLAM pipeline with a semantic OcTree to achieve robust localization and mapping while climbing the plant. To plan STEMbot's motion we propose a manifold-constrained A* planner along with ray-tracing goal specification to enable branch-aware traversal and the inspection of occluded targets. We validate our system through hardware experiments, demonstrating reliable traversal of stems ranging from 7-33mm and autonomous navigation across four distinct plant specimens. Quantitative evaluations show that our system achieves high-fidelity geometric reconstructions with an average Chamfer distance of less than 1cm relative to an offline photogrammetry baseline, confirming that STEMbot maintains the globally consistent odometry needed for autonomous navigation.
Zachary Charlick, Nilay Roy Choudhury, Haoyu Ma +2
Robotics Department, University of Michigan, Ann Arbor, MI, USA.