Design and Simulation-Based Evaluation of ARTgriculture 2.0, an IoT-Enabled Unmanned Ground Vehicle for Artistic Landscaping

As cities densify, urban green infrastructure and artistic landscaping, the deliberate arrangement of plants into visual compositions, carry ecological and public-health weight, yet manual execution does not scale. This article presents ARTgriculture 2.0, a redesigned Internet-of-Things-enabled unmanned ground vehicle that automates the full seeding cycle of digging, sowing, backfilling, and spraying for artistic landscaping, extending our earlier conference prototype. The platform introduces iris-seeding, variable-gap, and depth-control mechanisms sized and fabricated for compositional planting, a layered control architecture, and a browser-based interface for autonomous and manual operation. The redesigned mechanisms were fabricated and exercised on the physical chassis in indoor bench tests, without soil engagement and with no seed lot loaded, and no field trial has been performed; the validation level reported here is therefore module-level bench actuation plus system-level simulation. Full-system seeding is evaluated with a simulation model: a robot model built in NVIDIA Isaac Sim supplies the geometry and scene, while an offline kinematic execution model plans and executes area-fill coverage over regular grids and artistic patterns such as circles, diamonds, and hearts. Under the undisturbed model at a conservative low cruise speed, the platform reaches 96 to 100 percent seed placement at 2.4 to 3.4 cm mean tracking error and completes 15 m scale fields in 13 to 21 min; these are model values, not measurements. A Monte Carlo study of 30 runs per pattern that adds odometry noise, actuation-lag variation, wheel slip, lateral drift, and initial pose error holds mean coverage within 2.0 percentage points of the undisturbed values but roughly doubles mean tracking error, to 5.2 to 6.1 cm, and tightening the placement tolerance from 0.45 m to 0.15 m lowers coverage to 86.0 to 92.3 percent on the same trajectories, 90.1 to 92.3 percent in the reduced-speed mode. The functional prototype cost roughly 600 to 700 USD to build. The results indicate that pattern-precise, low-cost robotic seeding is achievable in simulation and identify soil-engaged field testing and quantified seed metering as the decisive next steps.