融合神经网络及麻雀算法的机器人避障研究

Research on Robot Obstacle avoidance based on neural network and Sparrow algorithm

摘要:

针对麻雀搜索算法（SSA）在机器人避障研究中，存在提早收敛于局部最优难以跳出、初始种群分布不够广泛、平衡能力差等问题对其进行改进。首先通过三层神经网络对规划环境进行栅格化建模；其次引入Halton序列得到初代种群分布，得到分布更广、更遍历的个体位置，提升后期寻优速度和效率；再次使用布朗运动优化麻雀位置更新的步长调节，帮助算法脱离局部优解，同时平衡全局切换局部的搜索节奏；最后，利用clothoid曲线法平滑路径，得到满足机器人机械性能的路径。经6个标准函数验证和Wilcoxon检验P值对比可知，改进后的算法相较于SSA和CSSA算法各项指标得到明显优化，且具有和SSA同一水平的时间复杂度。最后通过地图仿真得到平滑后的机器人避障路径。

Abstract:

Aiming at the problems of sparrow search algorithm (SSA) in the study of robot obstacle avoidance, such as early convergence to the local optimal and difficult to jump out, the initial population distribution is not wide enough, and the balance ability is poor, the improvement is made. Firstly, a three-layer neural network is used to model the planning environment. Secondly, Halton sequence was introduced to obtain the initial population distribution, and the individual locations with wider distribution and more traversal were obtained to improve the speed and efficiency of the later optimization. The Brownian motion was used to optimize the step size adjustment of sparrow position update, which helped the algorithm get rid of the local optimal solution and balanced the local search rhythm of global switch. Finally, the clothoid curve method is used to smooth the path, and the path satisfying the mechanical properties of the robot is obtained. After the verification of six standard functions and the comparison of Wilcoxon test P values, it can be seen that the improved algorithm is significantly optimized compared with SSA and CSSA algorithms in all indexes, and has the same level of time complexity as SSA. Finally, the smooth obstacle avoidance path of the robot is obtained through map simulation.