高遮挡温室场景下芦笋喷药机器人融合定位方法研究

Fusion Positioning Method of Asparagus Spraying Robot in High Occlusion Greenhouse Scene

摘要:

针对激光SLAM导航方案仍难以解决因作物冠层遮挡引发的定位误差逐步累积以及定位跳变等问题，尤其是在芦笋作物温室环境中，芦笋植株对传感器系统的遮挡使得机器人准确性受到严重影响。为解决这一问题，本文提出了一种改进Cartographer算法和融合超声标签定位技术的设施喷药机器人自主导航方法。通过IMU预积分获得增量位移、速度和姿态估计，引入滑动窗口优化策略，优化当前及历史数据状态估计。基于动态触发机制完成导航路径录制实现自主导航作业。在设施芦笋环境中进行了实际场景机器人定位轨迹映射和导航作业验证，试验结果表明，在速度0.2、0.4、0.6 m/s开展定位试验，本文方法定位轨迹和真实轨迹之间的平均绝对位姿误差分别为0.249、0.324、0.408 m，与Cartographer方法相比，平均定位误差降低7.4%、34.5%和30.6%;在速度0.2、0.4、0.6 m/s进行导航作业，机器人最大横向偏差为5.3 cm，当速度0.6 m/s时，平均横向定位误差为2.504 cm，机器人定位和导航精度均满足设施喷药机器人自主操作要求，为温室机器人自主作业提供了有效解决方案。

Abstract:

Navigation solutions based on laser simultaneous localization and mapping (SLAM) struggle to overcome the progressive accumulation of positioning errors and localization jumps caused by crop canopy occlusion. This issue was particularly severe in asparagus greenhouse environments, where the occlusion from asparagus plants significantly compromised the robot's accuracy. To address this challenge, an autonomous navigation method for a facility spraying robot was proposed, which integrated an improved Cartographer algorithm with ultrasonic tag positioning technology. The method obtained incremental displacement, velocity, and attitude estimations through inertial measurement unit (IMU) pre-integration and introduced a sliding window optimization strategy to refine state estimations of both current and historical data. A dynamic trigger mechanism was utilized to complete the recording of the navigation path, enabling autonomous operation. Field experiments involving robot localization trajectory mapping and navigation tasks were conducted in a real-world asparagus greenhouse. The results indicated that during positioning tests at speeds of 0.2 m/s, 0.4 m/s, and 0.6 m/s, the mean absolute pose errors between the proposed method's trajectory and the ground truth were 0.249 m, 0.324 m, and 0.408 m, respectively. Compared with the standard Cartographer method, these results represented a reduction in average positioning error by 7.4%, 34.5%, and 30.6%. During navigation tasks at the same speeds, the robot's maximum lateral deviation was 5.3 cm, and at 0.6 m/s, the average lateral positioning error was 2.504 cm. The positioning and navigation accuracy of the robot met the autonomous operational requirements for a facility spraying robot, providing an effective solution for autonomous operations in greenhouse environments.