結合環境探索策略與路徑規劃之適應計算性同時定位與建圖

Abstract

FastSLAM是目前解決同時定位與建圖最主要的方法，其中FastSLAM 2.0隨著地標的不斷增加，量測資訊與粒子內所存地標的比對次數也會大幅增加，導致計算效率降低。因此本論文提出一改良方法，稱之為「適應性計算之同時定位與建圖演算法(ACSLAM)」，在一開始的粒子更新階段係與FastSLAM 1.0相同，只採用里程計資訊，接下來在更新地標的階段，先選擇與量測資訊有最大相似性的地標先更新粒子狀態，再來更新地標。並且在重新取樣的階段使用「有效取樣大小」的值來決定下一次演算法的粒子數目，透過此方法來提高計算效率以及定位的精確度。然而單純運用SLAM演算法並無法進行環境探索與路徑規劃，因此本論文將ACSLAM整合基於邊緣偵測(frontier-based)之環境探索方法以及向量場路徑規劃，使機器人能完全自主性的執行任務。在實作方面，我們選擇了Pioneer 3-DX機器人作為移動平台，並搭配SICK感測器來偵測周圍環境，實驗結果證明，本方法可以使機器人在完全未知的環境下，自主地將環境探索完畢，並且完成建圖定位以及路徑規劃的任務。FastSLAM is a popular method to solve the Simultaneous Localization and Mapping (SLAM) problem. FastSLAM 2.0 adds the recent sensor measurement to improve the estimation accuracy compared to previous approaches. However, there is a runtime penalty when the number of landmarks becomes excessively large. To solve this problem, this thesis proposes a modified version for FastSLAM called adaptive computation SLAM (ACSLAM). In the beginning, ACSLAM only uses odometry information to estimate the robot’s pose. Particle state and landmark information are updated when a measurement has a maximum likelihood. To improve the computational efficiency, ACSLAM uses the effective sample size (ESS) to decide the number of particle for the next generation. The robot system is also extended with an exploration algorithm that uses the information generated by the SLAM system. By integrating the frontier-based exploration with ACSLAM and a path planning algorithm via potential field, the robot is capable of exploring an unknown environment safely in full autonomy. To practically evaluate the performance of the proposed method, a Pioneer 3-DX robot with a SICK laser scanner is used to validate the performance of the system both in simulation as well as practical experiments. Experimental results demonstrated that the proposed ACSLAM performed 40% faster than FastSLAM 2.0 with better accuracy.