Achieving autonomous navigation in mobile robotics requires the tight integration of perception, planning, and control within a unified on-board system. In off-road environments, fully closing the loop on-board with limited computational resources remains challenging because environmental uncertainty and terrain irregularities substantially raise the computational complexity of the required perception and control pipelines. This paper presents a cost-effective yet high-performance Unmanned Ground Vehicle (UGV) designed to bridge the gap between limited educational robots and expensive industrial platforms. The system adopts a dual-layer architecture in which an NVIDIA Jetson module executes neural components and the ROS2 Nav2 stack, while a dedicated real-time controller manages low-level locomotion. The two layers communicate over a CAN bus, ensuring robustness to electromagnetic interference and predictable communication latency. The platform is validated through an outdoor off-road navigation campaign involving trajectories with increasing geometric complexity. The experimental study shows the suitability of the proposed system for systematic evaluation of autonomous navigation performance in realistic off-road conditions.
An Integrated Edge-AI Architecture for Unstructured Off-Road Navigation
Adinolfi, Teodoro;Amato, Emilio
;Carletti, Vincenzo;Guarini, Alessio;Saggese, Alessia;Vento, Mario
2027
Abstract
Achieving autonomous navigation in mobile robotics requires the tight integration of perception, planning, and control within a unified on-board system. In off-road environments, fully closing the loop on-board with limited computational resources remains challenging because environmental uncertainty and terrain irregularities substantially raise the computational complexity of the required perception and control pipelines. This paper presents a cost-effective yet high-performance Unmanned Ground Vehicle (UGV) designed to bridge the gap between limited educational robots and expensive industrial platforms. The system adopts a dual-layer architecture in which an NVIDIA Jetson module executes neural components and the ROS2 Nav2 stack, while a dedicated real-time controller manages low-level locomotion. The two layers communicate over a CAN bus, ensuring robustness to electromagnetic interference and predictable communication latency. The platform is validated through an outdoor off-road navigation campaign involving trajectories with increasing geometric complexity. The experimental study shows the suitability of the proposed system for systematic evaluation of autonomous navigation performance in realistic off-road conditions.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
