AgriCruiser: Modular Agricultural Robotics

• AgriCruiser is a family of agricultural robotics platforms featuring low-cost construction, modular open-source design, and heterogeneous sensing for autonomous crop operations.
• The platform integrates diverse sensors—RGB stereo cameras, VIS–NIR line-scan, and thermal imagers—to build multi-layer maps for precise soil and crop analysis.
• Real-world testing demonstrates its effectiveness in precision weed management, adaptive navigation, and robust performance across varied field terrains.

AgriCruiser denotes a family of agricultural robotics platforms characterized by low-cost construction, modular open-source design, and heterogeneous sensing and actuation subsystems for autonomous ground mapping, navigation, and precision crop operations. Implementations span all-terrain unmanned ground vehicles (UGVs), over-the-row reconfigurable chassis, and hybrid drone-rover architectures, all targeting the automation of labor-intensive tasks such as soil characterization, weed management, and crop monitoring in structured and semi-structured field environments (Milella et al., 2021, Truong et al., 29 Sep 2025, Kant et al., 2023).

1. Platform Architectures and Mechanical Design

AgriCruiser platforms exhibit differentiated architectures depending on application scope and mobility requirements. The canonical over-the-row AgriCruiser employs an open-source, T-slot aluminum extrusion chassis with an adjustable track width of 1.42–1.57 m and 0.94 m chassis ground clearance, ensuring compatibility across diverse crop row geometries and growth stages. Propulsion is provided by dual front-driven wheels in a differential-drive configuration supported by two passive rear caster wheels, facilitating compact headland maneuvers with pivot-turn radii from 0.71 m to 0.79 m, experimentally matching R = W/2 for measured track widths (Truong et al., 29 Sep 2025). Continuous torque delivery (≥31.4 Nm per wheel), as validated by drive-train analyses, permits robust operation over variable terrain resistances and moderate inclines.

An alternate form factor incorporates a skid-steer UGV architecture (wheelbase 0.54 m × 0.70 m, ~0.8 m/s nominal speed), equipped with high-clearance chassis for traversal of unstructured ground surfaces and integration of multi-modal soil sensing payloads (Milella et al., 2021). In a hybrid UAV/UGV instantiation, a 650 mm Tarot quadrotor frame is modified to carry a skid-steer rover subsystem, employing four square-section aluminum rods for simultaneous flight landing gear and wheel support, achieving superior mechanical stiffness versus cylindrical supports (Kant et al., 2023).

Cost breakdowns for full-scale ground platforms are reported at $5,000–$6,000, primarily driven by COTS T-slot extrusions, powertrains, actuation, and subsystem hardware integration (Truong et al., 29 Sep 2025). Custom water-jet-cut adapters and minimal additional machining characterize the platform's manufacturability.

2. Sensor Integration and Multi-layer Environmental Mapping

Sensing modalities are central to AgriCruiser's autonomous and semi-autonomous capabilities. The canonical UGV platform incorporates:

• Exteroceptive sensing:
  • Calibrated RGB stereo cameras (Basler DART DaA1600-60uc, 1600×1200 px), performing dense depth estimation via semi-global matching.
  • VIS–NIR line-scan cameras (Specim V10, Basler ACA1920-155um, 400–1000 nm, 2 nm digital/8 nm optical), enabling computation of NDVI and spectral indices.
  • Thermal imagers (Micro-Epsilon ThermoImager160, 160×120 px, 16-bit), capturing surface radiance for soil/crop temperature profiling.
• Proprioceptive sensing:
  • 9-DOF IMU (x-IMU) with triaxial accelerometers, gyroscopes, and magnetometers (128 Hz), primarily for vehicle attitude and vibration response.

Sensors are precisely synchronized via a common timestamping scheme and jointly calibrated using checkerboard patterns (visible/thermographic) and extrinsic geometric alignment (via known pattern poses and physical overlap markers). The multi-layer map construction pipeline is as follows: 1. Stereo image rectification and depth extraction, 2. 3D point cloud reconstruction and rigid transformation to the vehicle frame, 3. Downsampling and statistical outlier removal, 4. Odometric stitching via visual odometry (libviso2), 5. Projection of 3D points to thermal and VIS–NIR imaging layers, aggregating radiometric, spectral, and geometric information, 6. Temporal alignment of proprioceptive acceleration data to ground patches under the vehicle (Milella et al., 2021).

This results in a multi-modal spatial map encoding metric, visual, spectral (e.g., NDVI), thermal, and mechanical (vibration) properties at each traversed ground patch.

3. Locomotion, Navigation, and Control Systems

Locomotion models span pure ground vehicles and hybrid platforms. The ground-based AgriCruiser employs differential drive kinematics with yaw rate $\omega = (v_r - v_l)/L$ and linear velocity $V = (v_r + v_l)/2$. Pivot turns (with one drive wheel stationary) are analytically and empirically validated to deliver minimal turning radii of $R = W/2$ for the given track width.

Mobile platforms utilize free-body analyses to determine required tractive forces and drive torques under various terrain resistances and inclines, ensuring overspecification of reported drive train (62.8 Nm per platform, operational usage <60%) (Truong et al., 29 Sep 2025).

For hybrid drone-rover variants, discrete state machine architectures mediate automatic mode transitions:

• Ground-Roll: Default straight-line navigation via skid steering.
• Obstacle-Evaluate: Obstacle detection (stereo depth threshold, typically 0.8 m).
• Weed-Action: Weeding (mechanical or spraying) upon weed recognition via depth and HSV color-space criteria.
• Drone-Flight: Autonomous vertical takeoff, forward traverse, landing, and resumption of ground navigation at obstacle-occluded locations (Kant et al., 2023).

Real-world deployments demonstrate robust mixed-terrain operation (concrete, gravel, grass, wet/dry soil), with torque utilization and speed scaling as functions of rolling resistance. Caster-lock modes reduce track misalignment in high-roughness conditions.

4. Precision Crop Operations and Experimental Performance

The AgriCruiser platform supports several crop-intervention subsystems.

Weed Management: Over-the-row AgriCruiser equipped with a 25-gallon tank, diaphragm pump (40 PSI), and four flat-fan nozzles (0.8 GPM total) achieved 19x–32x reductions in weed populations (pigweed, Venice mallow) in flax compared to manual weeding, with statistically significant reductions in both total weeds and crop damage ($p < 0.001$, two-sample t-test). Mean damaged plants per robotic-treated row were reduced 1.7x–5.9x versus manual (Truong et al., 29 Sep 2025).

Hybrid systems: Drone-rover variants implement mechanical weed plucking via a 3-DOF mini-arm (SG90 servos, effective up to 3 cm stem diameter, 90% removal success) or targeted spraying via a 5 V, 120 L/h submersible pump and single nozzle. The platform classifies green (“weed-like”) targets via RGB-HSV conversion and only actuates the relevant mechanism if classification is positive. Average mission time per obstacle/weed cycle is 27–35 s depending on row density (Kant et al., 2023).

Soil Characterization: AgriCruiser UGVs generate soil maps annotating patches (~0.85 × 0.70 m) with RGB-C1C2C3 moments, thermal statistics, NDVI indices, and root-mean-square vertical acceleration (mechanical roughness). One-sided CUSUM change detection algorithms operating on these features achieve high-precision, real-time transition detection (e.g., grass→pavement: precision ≈ 0.92, recall ≈ 0.95, RMS_az, 0.05–0.085 m/s² range) (Milella et al., 2021).

Operation	Performance Metric	Value(s)
Precision spraying	Weeds reduction vs. manual	19×–32×
Soil change detect	CUSUM (RMS_az, color, NDVI)—precision/recall	0.92/0.95 (grass→pavement)
Hybrid weed removal	Plucker success (stems ≤3cm); spraying success	90%; timed 5 s per weed
Terrain traversal	Torque utilization (wet soil); successful no-stall	57% (wet soil); all terrains

5. Open-Source Resources and Customization Pathways

Comprehensive design and implementation resources are openly available via repositories (e.g., https://github.com/StructuresComp/agri-cruiser) including full CAD assemblies, wiring diagrams, schematic files, ROS/ESP32 firmware (differential-drive, sprayer control, telemetry), and complete bills of materials with supplier references. Customization of the mechanical and electrical architecture is facilitated by the use of T-slot extrusions and off-the-shelf electronic/actuation subsystems (Truong et al., 29 Sep 2025).

Adaptation pathways include:

• Sensor payloads: RTK-GPS, LiDAR, multispectral cameras for phenotyping and advanced mapping.
• Actuator modules: precision seeding, mechanical or selective harvesting end-effectors.
• Suspension and terrain adaptation add-ons.
• Integration of full autonomy stacks, e.g., semantic segmentation-enabled navigation (AgroNav), variable-rate application, real-time terrain-type classification.

6. Limitations, Ongoing Research, and Future Directions

Reported limitations include the scanning nature of VIS–NIR line-scan cameras (single-row per frame, requiring odometric stitching), susceptibility to ambient lighting and thermal drift (necessitating periodic recalibration), and restricted plucker reach/force in hybrid systems. Hybrid drone-rover flight altitudes and trajectories are currently hard-coded, and real-time LIDAR-based mapping is absent. ROS-serial communication can introduce latency in modular controller setups.

Identified future work spans:

• Onboard, real-time semantic terrain classification.
• Extension to dynamic fluidic control for spraying while in motion.
• Expanded actuator payloads (e.g., torque and slippage sensors).
• Enhanced autonomy through online path replanning, height-aware flight (LIDAR), and swarm coordination for scalable field systems.
• Direct fusion of multi-layer maps into variable-rate precision agriculture platforms for adaptive seeding and fertilization (Milella et al., 2021, Truong et al., 29 Sep 2025, Kant et al., 2023).

A plausible implication is that AgriCruiser establishes a validated design pattern for cost-efficient, extensible agricultural automation platforms, enabling both academic research and practical deployment in small- to medium-scale settings through an open-source and modular ecosystem.