【作者】 朱刚；

【导师】 杨明；

【作者基本信息】 上海交通大学 ， 控制科学与工程， 2015， 硕士

【摘要】 磁导航技术由于其在可靠性方面的明显优势和广泛的应用场景而成为离实际应用最近的无人车技术。现有的一些演示应用项目都证明了磁导航方案的可行性。然而,在城市环境下,现有的磁导航方法在实现上遇到一些挑战,最主要的是城市环境下道路曲率较小,曲率突变较大,这使得磁传感器的无前瞻特性严重影响车辆的控制。现有的基于高精度磁钉坐标地图的方法可以解决这些问题,然而高精度磁钉坐标地图的建立过程非常繁琐,有时甚至不得不通过工程测绘方法一一测量。本文针对城市环境下磁导航无人车的核心问题和现有方法的不足,对磁导航技术进行了系统而深入的研究。首先,现有磁钉定位方法对传感器信息的利用不足,而基于磁场模型曲线拟合的方法又是非线性优化问题,计算复杂。针对这两个问题,本文提出了利用伪高斯曲线拟合法来实现磁钉的横向定位,将基于磁场模型曲线拟合的非线性优化问题化为伪高斯曲线拟合的线性最小二乘问题,并从数学上证明了二者在很高的精度上是可以相互近似的。另外,给出的弧长切向角坐标系下任意曲率可积道路模型的一般表达式,也为局部道路模型的建立提供了一般的解决方案。其次,现有基于高精度磁钉坐标地图的方法其地图生成过程比较复杂。针对这一问题,本文提出基于磁钉跟踪的曲率地图概念及其创建方法。和高精度磁钉坐标地图相比,曲率地图具有同等的前方道路描述作用,但其最大的优点是创建过程非常简单,手动驾驶车沿磁钉轨迹行驶一圈即可建立出来。最后,针对磁尺前瞻小,不利于车辆控制的问题,提出了基于曲率地图的磁导航预瞄控制方法。该方法利用任意曲率可积道路模型的一般表达式来融合磁钉跟踪结果信息和曲率地图信息,得到对传感器噪声和曲率地图误差鲁邦的前方道路信息,实现了磁导航无人车的预瞄控制。本文各个章节的实验部分都用实车实验数据验证了所提方法的有效性。实际中,本文所提出的基于曲率地图的磁导航方法取得了可以和基于高精度磁钉坐标方法相媲美的效果,却具有地图建立过程简单化,自动化的优点。更多还原

【Abstract】 Having the obvious advantages in reliability, magnetic navigation technology has broad application scenarios and has been the most applicable automated vehicle technology. Some of the existing demo applications have demonstrated the feasibility of magnetic navigation. However, in urban environment, the existing magnetic navigation approaches encountered some challenges. The most severe one is that the curvature of the road in urban environment is usually small and even varies violently, which makes the look-down property of magnetic navigated vehicle have heavily influence on its control. The existing magnetic nails’ high precision coordinates map based method can handle these problems, but the map’s acquisition is tedious, and sometimes nails’ coordinates need to be measured one by one through engineering of surveying. Aiming at the deficiencies of the existing methods and the key problems of magnetic navigation in urban environment, this paper make a profound study of magnetic navigation technology.First, the existing magnet localization methods don’t make full use of sensor’s information and the method of magnetic field model curve fitting is non-linear. To handle these deficiencies, this paper proposed the Gaussian curve fitting based approach for the lateral deviation’s calculation, using more sensor’s information and converting non-linear optimization of magnetic field model curve fitting to the least square problem of Gaussian curve fitting and verified mathematically the two curves are equivalent with very high accuracy. What’s more, this paper has proposed the normal model equation for any road whose curvature is integrable, thus provides normal solution for the building of local road model.Second, to overcome the problem that map generation of existing map based methods need high expenditure of time and efforts, the concept of curvature map and its generation algorithm were proposed. The proposed curvature map based approach could get the same performance with the approach based on nails’ high precision coordinates map, but its biggest advantage is its simple map generation process, which only needs the vehicle be driven along the reference trajectory by hand.Finally, to overcome the core issue of the look-down property of magnetic sensor, curvature map based navigation approach was proposed, in which the proposed local road model was used to fit the magnet tracking results as well as the curvature map information to achieve robust information of road ahead and thus preview control was achieved.In experimental section of each chapter, the proposed approaches have been verified using real car’s experimental data. In practical, the approach proposed in this paper performs as well as the approaches based on magnetic nails’ coordinates map. The curvature map using in our approach can be achieved simply and automatically.更多还原