基于多源信息融合的温室芦笋采收机器人自主导航控制方法

Autonomous Navigation and Control of Greenhouse Asparagus Picking Robot Based on Multi-sensor Information Fusion

摘要:

针对温室芦笋采收机器人在湿滑土壤环境中驱动系统易打滑、环境噪声干扰大,导致定位精度下降的问题,本文提出了一种融合IMU、里程计和激光雷达等多源传感器数据自主导航方法：在局部定位层面,设计引入打滑系数的扩展卡尔曼滤波（EKF）算法,实现对IMU与里程计数据的高精度融合;在全局定位层面,引入变异交叉机制以优化自适应蒙特卡洛定位（AMCL）算法,提升粒子滤波器全局搜索能力与鲁棒性。路径规划方面,提出基于目标方向耦合的改进人工势场法,克服传统方法易陷入局部极小的问题,提高路径规划的稳定性与环境适应性。试验结果表明,改进AMCL算法在x、y方向定位平均偏差分别为0.060、0.027 m,显著优于传统方法;EKF融合算法在x、y方向定位标准偏差分别为0.120、0.022 m,优于纯IMU与Odom定位方式。当行驶速度为0.1、0.2、0.3 m/s时,机器人路径跟踪平均横向和纵向偏差在0.121 m和0.096 m以内,最大航向角偏差不超过17.027°。该方法具备良好的定位精度,有效提升了机器人在湿滑环境中作业可靠性,能够满足温室内自主行走系统高精度建图、定位和导航需求。研究结果为采收机器人在农业环境中的行走应用提供了理论与技术支撑。

Abstract:

Aiming to address the challenges of localization accuracy degradation caused by slippage on wet soil and significant environmental noise in greenhouse conditions,an autonomous navigation method that fused data from multiple sensors was proposed,including IMU,odometry,and LiDAR. At the local localization level,an extended Kalman filter (EKF) algorithm incorporating a slippage coefficient was designed to achieve high-precision fusion of IMU and odometry data. At the global localization level,an adaptive Monte Carlo localization (AMCL) algorithm was optimized by using a mutation-crossover mechanism to enhance the global search capability and robustness of the particle filter. For path planning,an improved artificial potential field method based on target direction coupling was proposed to overcome the local minima problem of traditional methods,thereby improving path stability and environmental adaptability. Experimental results showed that the improved AMCL algorithm achieved localization standard deviations of 0.060 m in the x-direction and 0.027 m in the y-direction,significantly outperforming conventional methods. The EKF fusion algorithm achieved localization standard deviations of 0.120 m and 0.022 m in the x and y directions,respectively,surpassing standalone IMU and odometry-based localization. At three driving speeds (0.1 m/s,0.2 m/s,and 0.3 m/s),the average lateral and longitudinal tracking errors were maintained within 0.121 m and 0.096 m,respectively,with a maximum heading angle deviation of no more than 17.027°. The proposed method demonstrated good localization accuracy and path tracking performance,significantly improving the operational reliability of the robot in slippery environments. It met the requirements of high-precision mapping,localization,and navigation for greenhouse autonomous walking systems and provided theoretical and technical support for the deployment of harvesting robots in agricultural environments.