School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510640, China
Show less
文章历史+
Received
Published
Revised
2025-03-26
2026-03-18
2025-12-10
Online Date
2026-06-26
PDF (11695K)
摘要
近年来,基于深度强化学习(deep reinforcement learning, DRL)的方法在采摘路径规划领域受到广泛关注,但是目前的研究只单独考虑采摘机械臂,忽视了移动底盘与机械臂之间紧密相连的关系,导致不能充分发挥采摘机器人兼顾移动性与操作性的优势。为此,提出一种基于改进双延迟深度确定性策略梯度算法(twin delayed deep deterministic policy gradient, TD3)的荔枝采摘机器人协同路径规划方法。首先,通过深度强化学习对采摘机器人协同路径规划问题进行求解,根据协同路径规划任务设置状态空间与动作空间,并综合考虑位置、姿态与避障任务设计复合目标奖励函数;其次,引入双Actor网络与层归一化对TD3算法进行改进;最后在CoppeliaSim仿真平台内搭建非结构化荔枝采摘环境,对采摘机器人进行训练。仿真实验结果显示,深度强化学习能较好地处理协同路径规划任务,改进后的TD3算法收敛速度提升,在测试实验中路径规划成功率达96%,比原始算法提高了近69.9个百分点。结果表明:本系统能够较好地解决采摘机器人协同路径规划引起的高维度问题,能让移动底盘与机械臂进行协同规划,有助于提高采摘机器人的智能化水平,并在采摘路径规划成功率方面有一定的优势,可为采摘机器人路径规划的未来研究方向提供参考。
Abstract
In recent years, methods based on deep reinforcement learning (DRL) have received significant attention in the field of picking path planning. However, existing research is primarily focused on the picking manipulator, while the close coordination between the mobile chassis and the manipulator is often overlooked. As a result, the full potential of the robot in terms of mobility and manipulation is not fully utilized. To address this limitation, a collaborative path planning method for litchi picking robot was proposed using an improved twin delayed deep deterministic policy gradient (TD3) algorithm. Initially, the collaborative path planning problem was modeled using deep reinforcement learning. The state and action spaces were defined according to the requirements of the task, and a composite reward function was formulated, taking into account position, orientation, and obstacle avoidance. Subsequently, the TD3 algorithm was enhanced through the integration of dual Actor networks and layer normalization. Finally, an unstructured litchi picking environment was developed on the CoppeliaSim simulation platform, and the robot was trained within this simulated environment. The simulation results show that deep reinforcement learning effectively addresses the collaborative path planning problem. The improved TD3 algorithm achieves faster convergence. In the testing phase, the path planning success rate reaches 96%, representing an improvement of nearly 69.9 percentage points over the original algorithm. These results indicate that the proposed system effectively overcomes the challenges posed by high-dimensional state-action spaces in collaborative path planning. It enables coordinated motion between the mobile chassis and the manipulator, thereby enhancing the robot’s operational intelligence. Furthermore, it demonstrates a clear advantage in planning success rate and offers valuable guidance for future research in picking robot path planning.
LI Zhen-dong, HU Guang-hua.
Collaborative Path Planning of Litchi Picking Robot Based on Improved TD3 Algorithm[J].
Science Technology and Engineering, 2026, 26(8): 3367-3376 DOI:10.12404/j.issn.1671-1815.2502051
ChenHoubin, YangShengnan, SuZuanxian, et al. Analysis of the national litchi production in 2024 and management suggestions[J]. China Tropical Agriculture, 2024(3): 8-20.
XiongJuntao, LiZhongxing, ChenShumian, et al. Virtual robot picking path obstacle avoidance planning based on deep reinforcement learning[J]. Transactions of the Chinese Society for Agricultural Machinery, 2020, 51(S2): 1-10.
[7]
HartP E, NilssonN J, RaphaelB. A formal basis for the heuristic determination of minimum cost paths[J]. IEEE Transactions on Systems Science and Cybernetics, 1968, 4(2): 100-107.
[8]
TangZ, XuL, WangY, et al. Collision-free motion planning of a six-link manipulator used in a citrus picking robot[J]. Applied Sciences, 2021, 11(23): 11336.
LiXiaojuan, ChenTao, HanRuichun, et al. Research on path planning of robotic arm with improved RRT algorithm in uncertain environment for harvesting[J]. Journal of Chinese Agricultural Mechanization, 2024, 45(4): 193-198.
GongLixiong, ChenJialin, HuangXiao, et al. Aircraft deck flat path planning based on improved Informed-RRT*algorithm[J]. Science Technology and Engineering, 2024, 24(17): 7429-7437.
[13]
ZhuangM, LiG, DingK. Obstacle avoidance path planning for apple picking robotic arm incorporating artificial potential field and A* algorithm[J]. IEEE Access, 2023, 11: 100070-100082.
[14]
唐座亮. 柑橘采摘机械臂运动规划研究[D]. 雅安: 四川农业大学, 2022.
[15]
TangZuoliang. Research onmotion planning of the citrus-picking manipulator[D]. Yaan: Sichuan Agricultural University, 2022.
[16]
LiuC, FengQ, TangZ, et al. Motion planning of the citrus-picking manipulator based on the TO-RRT algorithm[J]. Agriculture, 2022, 12(5): 581-603.
ZhangYongmei, ZhaoJiarui, WuAiyan. Robot path planning algorithm based on curiosity-driven deep reinforcement learning[J]. Science Technology and Engineering, 2022, 22(25): 11075-11083.
WangWei, WuHao, LiuHongxun, et al. An attitude controller for quadrotor drone using RM-DDPG[J]. Science Technology and Engineering, 2023, 23(34): 14888-14895.
[21]
FujimotoS, HoofH, MegerD. Addressing function approximation error in actor-critic methods[C]// International Conference on Machine Learning. Stockholm: PMLR, 2018: 1587-1596.
[22]
LinG, ZhuL, LiJ, et al. Collision-free path planning for a guava-harvesting robot based on recurrent deep reinforcement learning[J]. Computers and Electronics in Agriculture, 2021, 188: 106350.
YangShuhua, XieXiaobo, BingZhenkai, et al. Path planning of green walnut picking robotic arm based on HER-TD3 algorithm[J]. Transactions of the Chinese Society for Agricultural Machinery, 2024, 55(4): 113-123.
ZhengChange, GaoPo, GanHao, et al. Trajectory planning method for apple picking manipulator based on stepwise migration strategy[J]. Transactions of the Chinese Society for Agricultural Machinery, 2020, 51(12): 15-23.
LiYong, ZhangChaoxing, ChaiLiaoning. Collaborative obstacle avoidance trajectory planning for mobile robotic arms based on artificial potential field DDPG algorithm[J]. Computer Integrated Manufacturing Systems, 2024, 30(12): 4282-4291.
[29]
WangC, ZhangQ, TianQ, et al. Learning mobile manipulation through deep reinforcement learning[J]. Sensors, 2020, 20(3): 939.
[30]
SuttonR S, BartoA G. Reinforcement learning: an introduction[M]. Cambridge: MIT Press, 2018: 1-13.
[31]
ShakyaA K, PillaiG, ChakrabartyS. Reinforcement learning algorithms: a brief survey[J]. Expert Systems with Applications, 2023, 231: 120495.
[32]
MnihV, KavukcuogluK, SilverD, et al. Human-level control through deep reinforcement learning[J]. Nature, 2015, 518(7540): 529-533.
[33]
CiosekK, VuongQ, LoftinR, et al. Better exploration with optimistic actor critic[J]. Advances in Neural Information Processing Systems, 2019, 32.
[34]
LüJ, MaX, YanJ, et al. Efficient continuous control with double actors and regularized critics[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Vancouver: AAAI, 2022, 36(7): 7655-7663.
[35]
YueY, LuR, KangB, et al. Understanding, predicting and better resolving Q-value divergence in offline-RL[J]. Advances in Neural Information Processing Systems, 2023, 36: 60247-60277.
[36]
ZhouQ, LianY, WuJ, et al. An optimized Q-learning algorithm for mobile robot local path planning[J]. Knowledge-Based Systems, 2024, 286: 111400.
[37]
WangZ, HuntJ J, ZhouM. Diffusion policies as an expressive policy class for offline reinforcement learning[J]. arXiv Preprint, arXiv: 220806193, 2022.
[38]
LiuH, YingF, JiangR, et al. Obstacle-avoidable robotic motion planning framework based on deep reinforcement learning[J]. IEEE/ASME Transactions on Mechatronics, 2024, 29(6): 4377-4388.
[39]
KartiwaA, SupriatnaA K, RusyamanE, et al. Review of quaternion differential equations: historical development, applications, and future direction[J]. Axioms, 2023, 12(5): 483.