• Developed an autonomous grapevine-pruning robot integrating vine-structure estimation, motion planning, and end-effector visual correction. • Motion planning was adapted to dynamically changing vine structures through obstacle-model updates after each cutting operation. • High-resolution close-range observations from the end-effector camera significantly improved pruning accuracy and robustness. • Achieved practical field performance with an 82% pruning success rate and reliable bud-count control. This study presents an autonomous robotic pruning system for grapevines that integrates vine-structure estimation, motion planning, and a vision-based end-effector position-correction mechanism. Field trials on 16 vines (10 single-pass; 6 two-stage) achieved pruning success rates of up to 82% and completed pruning of a single vine in 92 s under our test conditions. Enabling the position-correction module increased the per-task success rate from 68% to 82%; this difference was statistically significant in our single-pass analysis (Mann–Whitney U = 90.5, p = 0.040). The correction step itself succeeded in 90% of attempts and added 0.98 s (8.8%) to the cycle time. Bud-count control met or modestly exceeded the target in >85% of successful cuts. Comparing operational strategies at the set level, the two-stage approach showed a higher task success rate and fewer collisions than single-pass pruning (Mann–Whitney tests), while other failure modes were broadly similar. These findings indicate promising but preliminary performance. The evaluation is limited in scale (n = 16 vines), conducted in a single season, and focuses on one platform and vineyard context; consequently, generalizability remains to be established. Future work will address retry mechanisms, depth-sensing stability, and throughput improvements via pruning-order optimization and dual-arm operation.
Hizatate et al. (Sun,) studied this question.