文献类型： 外文期刊

作者： Li, Yajun ¹ ; Feng, Qingchun ¹ ; Zhang, Yifan ¹ ; Peng, Chuanlang ¹ ; Ma, Yuhang ¹ ; Liu, Cheng ¹ ; Ru, Mengfei ¹ ; Sun, Jiahui ¹ ; Zhao, Chunjiang ¹ ;

作者机构： 1.Beijing Acad Agr & Forestry Sci, Intelligent Equipment Res Ctr, Beijing 100097, Peoples R China

2.Hunan Agr Univ, Coll Mech & Elect Engn, Changsha 410128, Peoples R China

3.Beijing Key Lab Intelligent Equipment Technol Agr, Beijing 100097, Peoples R China

4.Beijing Acad Agr & Forestry Sci, Intelligent Equipment Res Ctr, 11 Shuguang Garden Middle Rd, Beijing 100097, Peoples R China

关键词： Collision-free; Deep reinforcement learning; Opimal operating posture; Tomato-harvesting robot

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：8.3； 五年影响因子：8.3 ）

ISSN： 0168-1699

年卷期： 2024 年 216 卷

页码：

收录情况： SCI

摘要： Collision-fros grasping of the thin, brief peduncies connecting sherry tomate clusters to the main stem was crucial for tomato harvesting robots. Recognising that the optimal operating posture for each individual poduncle was various, this study proposed a naval pedumele grasping posture decision model using deep rein forcement learning (DRL) for tomato harvesting manipulators, to ove is the collision issue caused by Axed- posture grasping. This model could dynamically generated action sequences for the harvesting manipulator, ensuring that the end-effectar approach to the peduncle along the collision-fros path with the optimal grasping posture. Building upon price research inte the multi-task identification of tomato clusters, peduncles, and the main stom, a keypoint-based spatial pose description model for tomate bunches was devised. Through this, the optimal operating pesture for the and effector on the peduncle was established. An improved HER-SAC (Soft Actor Critic with Hindsight Experience Replay) algorithm was subsequently established to guide the and-effector in collision-free grasping motions. The reward function of this algorithm incorporated end-effccser posture constraints obtained from the optimal posture plans. In the training phase, a heuristic strategy model, providing prior knowizdgs, was marged with a dynamic guin module to sidestep local optimal policies, collectiv enhancing the learning efficiency In the simulation, our method improved the success rate of the peduncle grasping by at least 14 %, compared with SAC, HER DOFC and HER-TD3. For the identical scenarios, improved HER-SAC reached the desired posture with a minimum of 15.5 % fewer stops compared to ather algorithms. In Feld cxperiments conducted in tomate greenhouses, the robot schieved a harvesting success rate of 35.5 which was an increase of 57.31% and 43.0 1% compared to traditional methods with food horizontal and parallel to-main-stem postures, respectively. The average operation time, from identification to successful harvesting, was 11.42 Our findings offer a promising solution is enhancing the efficiency of tomato-harvesting robots.