A novel navigation system using driving state classification algorithm for semi-autonomous tractors

초록

The rapidly growing global population, migration, and urbanization are posing threats to the sustainability of agriculture and food security. Consequently, an autonomous driving technology for the agricultural machinery is gaining significant attention as a potential solution to these challenges. Semi-autonomous technologies such as auto-steering systems are being researched and developed. The auto-steering system has relatively low initial installation costs while still enables a tractor to automatically and precisely follow a predetermined path. However, the existing navigation systems for auto-steering system have limitations, such as constraints on initial motion for navigation system alignment, challenges in driving state identification, and excessive reliance on GPS. This paper proposes a novel navigation system integrated with a LSTM-based classification algorithm and adaptive Kalman filter, which can be applied to semi-autonomous tractors using auto-steering systems. The LSTM-based classification algorithm uses IMU data to predict whether the machinery is in stationary state, backward, or forward movements. The proposed navigation system allows the semi-autonomous tractor to perform the initial alignment during linear forward and backward motion and to apply zero velocity update effectively when the GPS signal is weak or even unavailable. The proposed algorithms were validated through extensive experiments using a tractor on both roads and farm fields. The experimental results showed that the proposed classification algorithm outperformed conventional zero velocity detectors and other machine learning methods in terms of robustness, accuracy, and versatility. The proposed navigation system experimentally demonstrated its practical applicability.

키워드