基于模糊K均值聚类和Sarsa(λ)算法的自适应爬壁机器人路径规划

Adaptive Path Planning of Wall-Climbing Robot Based on MIP and Improved Fuzzy K-Means Algorithm and Sarsa(λ)

摘要:

针对现有爬壁机器人规划算法难以实现在线自适应高效规划的问题,设计了一种基于模糊K-Means算法和经典Sarsa(λ)算法自适应爬壁机器人规划算法；首先,对爬壁机器人的动力学模型进行了建模和分析,然后,对爬壁机器人规划中的状态进行自适应聚集从而实现值函数的近似,设计了K值可变的改进模糊K均值聚类算法对状态进行自适应地在线聚类,将聚类中心对应的值函数作为整个聚类所有数据对象的值函数的近似值,最后,对基于模糊K均值聚类算法和Sarsa(λ)算法的爬壁机器人在线规划算法进行了定义和描述,在MATLAB环境下对简单障碍物场景和复杂障碍物场景分别仿真实验,实验结果表明文中方法能有效地进行路径规划,随着情节数的增加,规划结果逐渐收敛到最优值,同时在环境变化时,收敛效果不受影响,具有较好的稳定性,是一种高效地实现爬壁机器人在线规划的方法。

Abstract:

Aiming at the given wall-climbing robot path planning algorithm difficult to plan effectively online, an adaptive path planning algorithm based on K-Means algorithm and classic Sarsa(λ) algorithm are designed. Firstly, the dynamical model for wall-climbing robot is designed. Then the states of the planning space is clustered adaptively to realize the value function approximating, the improved fuzzy K-Means with the variable K is designed to cluster on-line, using value of the cluster center as the approximate value for all the sample of the whole cluster. Finally, the algorithm based on fuzzy K-Means and classic Sarsa(λ) algorithm for wall-climbing robot path planning is defined and described. The simulation experiment with simple barrier and complicate barrier is operated in the MATLAB, the result shows the method in this paper can realize the path planning, and with the increase of the episode, the planning result is converged to the optimal value, and also the convergence effect is not subject to the change of the environment, so it has strong feasibility. It is a on-line planning method for wall-climbing robot path planning.