Answers are grounded in this paper and cite their sources.
Mechanized main crop harvesting in ratoon rice requires joint optimization of header coverage, stubble crushing, and travel distance. Under common track and header geometry constraints, 12 candidates combining four path patterns and three turning types were compared. The highest-scoring candidate provided a baseline for deep reinforcement learning of local driving and channel transitions under a fixed channel sequence, followed by field comparison. The boustrophedon route combining U-turns and fishtail turns achieved the highest composite score. Relative to this baseline, the deep Q-network (DQN) route increased header-covered area by 7.17%, shortened travel distance by 22.17%, and reduced turns from 55 to 37. Among four DQN-based algorithms, Double DQN (DDQN) achieved favorable rewards and path metrics. In field trials, conventional and DDQN routes had mean single-pass crushing rates of 36.35% and 31.83%, and repeated-pass rates of 15.85% and 12.77%, respectively. Under the tested conditions, local policy learning combined with whole-field trajectory assembly showed potential to improve coverage and reduce crushing. The method informs rule-constrained harvest path design; its effects on ratoon crop yield require further evaluation.
No takes yet. Share an insight, caveat, or question.
Lin et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: