模块化机器人最优越野构型神经网络规划方法

doi:10.12382/bgxb.2023.0713

摘要/Abstract

摘要：

轮式模块化机器人在满足人类对于无人自主任务需求时具有很多优势,机器人组合体构型在搬运物资、山地越障等方面更具有独特优势,为此提出一种多模块机器人构型优化规划方法。构建数字化地形表达,建立参数化地形辨识模型,运用遗传算法构建能耗与时间加权组合的最优构型,改变约束条件在不同地形下进行大量平行运行得到大量地形-最优构型参数结果对,将地形集合构建为输入集,将最优构型集合构建为输出集,训练借助神经网络技术快速得到面向任意地形的最佳组合体构型,使得组合体在面对三维复杂地形时实现高成功率、高可靠性越障运动,同时将能耗成本和时间成本降至最低。通过物理引擎平台仿真搭建仿真野外地形,对规划得到的构型进行通过性验证和性能测试,各构型均能完成地形跨越,同时验证规划算法的优化能力;搭建模块化机器人样机实物进行实验,以6×1刚性连接构型完成了2倍轴距宽沟壑的跨越。研究结果表明,所提方法能够高效地规划各类地形下满足通过性要求和时间能耗最优的组合体越障构型。

关键词: 轮式机器人, 山地越障, 构型优化, A^*算法, 遗传算法, 反向传播神经网络

Abstract:

Wheeled modular robots have many advantages in meeting human needs for unmanned autonomous tasks, and the robot combination configuration has unique advantages in materials handling, mountain obstacle crossing, and other aspects. Therefore, a multi-module robot configuration optimization planning method is proposed. A digital terrain representation is constructed, and a parameterized terrain identification model is established. And then the genetic algorithm is used to construct an optimal configuration of energy consumption and time weighted combination, change the constraint conditions and perform a large number of parallel runs under different terrains to obtain a large number of terrain optimal configuration parameter pairs. The terrain set is constructed as the input set, and the optimal configuration set is constructed as the output set. The optimal combination configuration for any terrain is quickly obtained by training with neural network technology, so that the combination can achieve high success rate and high reliability obstacle crossing motion when facing a three-dimensional complex terrain, while minimizing the energy and time costs. A simulated field terrain is simulated and built through the physics engine platform, the feasibility and performance of the planned configurations are verified, ensuring that each configuration can complete terrain crossing. At the same time, the optimization ability of the planning algorithm is verified. A modular robot prototype is tested, and the crossing of a 2-times wheelbase wide ravine is completed by using a 6×1 rigid connection configuration. The research results show that the proposed method can be used to efficiently plan the combination obstacle crossing configuration that meets the requirements of passability, and optimal time and energy consumption in various terrains.

Key words: wheeled robot, mountain obstacle crossing, configuration optimization, A^* algorithm, genetic algorithm, BP neural network

中图分类号:

TP242.3

党婉莹, 周乐来, 李贻斌, 张辰. 模块化机器人最优越野构型神经网络规划方法[J]. 兵工学报, 2024, 45(10): 3674-3685.

DANG Wanying, ZHOU Lelai, LI Yibin, ZHANG Chen. Neural Network Planning Method for Optimal Off-road Configuration of Modular Robots[J]. Acta Armamentarii, 2024, 45(10): 3674-3685.

图/表 22

图1 系统构型优化实现流程

Fig.1 Flowchart of system configuration optimization

图2 四轮差速机器人运动模型示意图

Fig.2 Schematic diagram of motion model of a four-wheel differential robot

图3 构型转弯示意图

Fig.3 Schematic diagram of configuration turning

表1 输入集参数

Table 1 Input set parameters

参数	数值
沟壑宽度/m	3
横向最大坡度/(°)	5
最大可通行宽度/m	1
纵向最大上升坡度/(°)	15

表2 输出集参数表

Table 2 Output set parameters

输出集	表达式
1	$p 1 e 1$
2	$p 2 e 2$
3	1,2

表2 输出集参数表

Table 2 Output set parameters

输出集	表达式
1	$p 1 e 1$
2	$p 2 e 2$
3	1,2

表3 机器人模块参数

Table 3 Parameters of robot module trolley

参数	参数
车身尺寸(长×宽×高)/m	0.7×0.3×0.6
对接机构总长度/m	0.1
车体质量/kg	30
离地间隙/m	0.55
接近角/(°)	75
离去角/(°)	60
运动速度/(m·s^-1)	0.5

图4 以六模块机器人为例的6×1构型仿真示意图

Fig.4 Simulation diagram of a 6×1 configuration using a six- module robot

图5 3组预测值与真实值间误差

Fig.5 Error values among three sets of predicted and true values

图6 侧斜坡寻路示意图

Fig.6 Schematic diagram of side slope path finding

图7 侧斜坡仿真模型

Fig.7 Side slope simulation model

图8 2×3刚性构型越障过程

Fig.8 Obstacle crossing process for 2×3 rigid configuration

图9 构型z轴坐标变化

Fig.9 Configuration z coordinate change curve

图10 6×1刚性构型对比实验侧向倾覆

Fig.10 State diagrams of lateral overturning in a 1×6 rigid configuration comparative experiment

图11 高台连绵地形寻路示意图

Fig.11 Schematic diagram of path finding in high terrain and continuous terrain

图12 高台连绵地形结合示意图

Fig.12 Schematic diagram of the combination of high terrain and continuous terrain

图13 6×1柔性构型越障过程

Fig.13 State diagram of obstacle crossing process for 6×1 flexible configuration

图14 构型z轴坐标变化

Fig.14 Configuration z coordinate change curve

图15 对比实验

Fig.15 Contrast test

表4 优化性对比仿真结果

Table 4 Optimization versus simulated results

构型	通过性	能耗/J	时间/s
1×6刚性	×	×	×
1×6柔性	×	×	×
2×3刚性	√	4837.5	43
2×3柔性	√	4837.5	43
3×2刚性	√	5017.5	44.6
3×2柔性	√	5017.5	44.6
4×3刚性	√	5197.5	46.2
4×3柔性	√	5197.5	46.2
5×2刚性	√	5377.5	47.8
5×2柔性	√	5377.5	47.8
6×1刚性	×	×	×
6×1刚性	×	×	×

表5 麦轮平台参数

Table 5 Mecanum wheel platform parameters

参数	数值
车身尺寸(长×宽×高)/m	0.22×0.26×0.2
对接机构总长度/m	0.07
车体质量/kg	2.9
离地间隙/m	0.05
接近角/(°)	69
离去角/(°)	78
运动速度/(m·s^-1)	1.0

图16 6×1刚性构型越沟壑过程

Fig.16 State diagram of 6×1 rigid configuration crossing a gully

表6 遗传算法运算与神经网络运算时间

Table 6 Neural network operation times

序号	遗传算法	神经网络
1	2.77	1.41
2	3.402	1.58
3	2.46	1.09

参考文献 22

[1]	沙莎, 王辉平. 轮式移动机器人滑移轨迹跟踪控制策略研究[J]. 机床与液压, 2023, 51(9): 62-69.
	SHA S, WANG H P. Research on sliding trajectory tracking control strategy for wheeled mobile robots[J]. Machine Tool & Hydraulics, 2023, 51(9): 62-69. (in Chinese)
[2]	刘松, 柴汇, 李贻斌, 等. 电动力液压驱动四足双臂机器人的设计与实现[J]. 机器人, 2022, 44(6): 649-659.
	LIU S, CHAI H, LI Y B, et al. Design and implementation of electro-hydraulically driven quadruped two-arm[J]. Robot, 2022, 44(6): 649-659. (in Chinese)
[3]	SHANG H, XI Y R, XIE Z B, et al. A configuration design method for reconfigurable robots based on terrain understanding[J]. Journal of Physics: Conference Series, 2022, 2203: 012023.
[4]	OETOMO D, DANEY D, HARADA K, et al. Topology design of surgical reconfigurable robots by interval analysis[C]// Proceedings of 2009 IEEE International Conference on Robotics and Automation. Kobe, Japan:IEEE, 2009: 3085-3090.
[5]	KUROKAWA H, TOMITA K, KAMIMURA A, et al. Distributed self-reconfiguration of M-TRAN III modular robotic system[J]. The International Journal of Robotics Research, 2008, 27(3/4): 373-386.
[6]	ZHAO J, TANG S F, ZHU Y H. Design and implementation of a modular self-reconfigurable robot[J]. High Technology Letters, 2009, 15(3): 227-232.
[7]	夏平, 朱新坚, 费燕琼. 一种新型自重构模块机器人的设计和运动策略[J]. 中国机械工程, 2006(15): 1549-1552.
	XIA P, ZHU X J, FEI Y Q. Design and motion strategy of a novel self reconfigurable modular robot[J]. China Mechanical Engineering, 2006 (15): 1549-1552. (in Chinese)
[8]	费燕琼, 夏振兴, 夏平. 自重构机器人的基本模块结构设计与分析[J]. 中国机械工程, 2007(9): 1085-1088.
	FEI Y Q, XIA Z X, XIA P. Basic module structure design and analysis of self reconfigurable robots[J]. China Mechanical Engineering, 2007(9): 1085-1088. (in Chinese)
[9]	杨振, 付庄, 管恩广, 等. M-Lattice模块机器人的运动学分析及构型优化[J]. 上海交通大学学报, 2017, 51(10): 1153-1159. doi: 10.16183/j.cnki.jsjtu.2017.10.001
	YANG Z, FU Z, GUAN E G, et al. Kinematic analysis and configuration optimization of M-Lattice modular robots[J]. Journal of Shanghai Jiao Tong University, 2017, 51(10): 1153-1159. (in Chinese)
[10]	WU W Q, GUAN Y S, YANG Y F, et al. Multi-objective configuration optimization of assembly-level reconfigurable modular robots[C]// Proceedings of the 2016 IEEE International Conference on Information and Automation (ICIA). Ningbo, China: IEEE, 2016: 528-533.
[11]	LIU L S, WANG B, XU H. Research on path-planning algorithm integrating optimization A-star algorithm and artificial potential field method[J]. Electronics, 2022, 11(22): 3660.
[12]	GANGANATH N, CHENG C T, TSE C K. A constraint-aware heuristic path planner for finding energy-efficient paths on uneven terrains[J]. IEEE Transactions on Industrial Informatics, 2015, 11(3): 601-611.
[13]	高九州, 徐威峰, 张立辉, 等. 基于改进A^*算法的无人机避障航线规划[J]. 现代电子技术, 2023, 46(8): 181-186.
	GAO J Z, XU W F, ZHANG L H, et al. UAV obstacle avoidance route planning based on improved A^* algorithm[J]. Modern Electronic Technology, 2023, 46(8): 181-186. (in Chinese)
[14]	陈春良, 齐鸥, 魏兆磊, 等. 基于蒙特卡洛仿真和遗传算法的车辆装备保障运输网络优化[J]. 兵工学报, 2016, 37(1): 114-121. doi: 10.3969/j.issn.1000-1093.2016.01.018
	XU C L, QI O, WEI Z L, et al. Optimization of vehicle equipment support transportation network based on Monte Carlo simulation and genetic algorithm[J]. Journal of Ordnance Engineering, 2016, 37(1): 114-121. (in Chinese)
[15]	AN X B, HU C, LI Z H, et al. Decentralized AdaBoost algorithm over sensor networks[J]. Neurocomputing, 2022, 479: 37-46.
[16]	OLUWASEUN O M, ADEYEMI A A, MOSES O O, et al. An improved multi-objective a-star algorithm for path planning in a large workspace: design, implementation, and evaluation[J]. Scientific African, 2021, 15: e01068.
[17]	JIN G Y, FENG W M, QING P. Prediction of waterway cargo transportation volume to support maritime transportation systems based on GA-BP neural network optimization[J]. Sustainability, 2022, 14(21): 13872.
[18]	GARCIA J C S, TANAKA H, GIANNETTI N, et al. Multiobjective geometry optimization of microchannel heat exchanger using real-coded genetic algorithm[J]. Applied Thermal Engineering, 2022, 202: 117821.
[19]	GUNTER R. Convergence analysis of canonical genetic algorithms[J]. IEEE Transactions on Neural Networks, 1994, 5(1): 96-101. pmid: 18267783
[20]	ELHASSANIA M. Solving a stochastic programming with recourse model for the stochastic electric capacitated vehicle routing problem using a hybrid genetic algorithm[J]. European Journal of Industrial Engineering, 2022, 16(1): 71-90.
[21]	ZHANG Z, LIU S Y, ZHANG Y Q, et al. Optimization of low-power femtosecond laser trepan drilling by machine learning and a high-throughput multi-objective genetic algorithm[J]. Optics & Laser Technology, 2022, 148: 107688.
[22]	ZHANG Z J, SUN R, XU T, et al. Robot path planning based on shuffled frog leaping algorithm combined with genetic algorithm[J]. Journal of Intelligent & Fuzzy Systems: Applications in Engineering and Technology, 2023, 44(3): 5217-5229.