Design and Experimental Validation of Modular Autonomous Driving Platform for Off-Road 8 × 4 Trucks in Agricultural Operation

This study introduces and experimentally validates an autonomous driving platform for off-road <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$8\times 4$ </tex-math></inline-formula> heavy-duty trucks, aimed at sugarcane harvesting operations. The system integrates a minimally invasive drive-by-wire interface for throttle, braking, and steering; a GNSS-based localization module; and a lightweight perception suite comprising a 2D LiDAR and a monocular camera. A modular ROS2-based architecture supports real-time perception, planning, and control. The platform was tested on an off-road track in Brazil across eight driving scenarios, including straight paths, curves, and obstacle avoidance at varying speeds. A spatial evaluation framework was implemented using arc-length reparameterization and KD-Tree nearest-neighbor matching to compare actual and reference trajectories. Metrics included lateral and heading RMSE, IAE, ITAE, jerk, and obstacle clearance. Results showed consistent high lateral tracking accuracy, with RMSE reaching 0.15 m in straight segments and remaining below 1.0 m in most complex scenarios. Scenario T8, involving curvature and obstacle deviation, surpassed this threshold due to increased control demands. Across 40 runs, heading RMSE remained below 10°, obstacle clearance above 1.5 m, and jerk below 0.5 m/s3. Geometrically modeled references yielded more stable heading control, while manually recorded paths resulted in smoother dynamic profiles. Statistical analysis confirmed that increased trajectory complexity degrades performance. Scenario T6 outperformed others in lateral stability, though results were statistically like T1 under simpler conditions. The platform demonstrated reliable tracking and real-time operation in scenarios representative of off-road agricultural tasks. Future work will focus on mitigating GNSS latency, improving post-obstacle recovery, and integrating sensor redundancy and adaptive planning to enhance deployment in agricultural logistics.