【作者】 许烁；

【导师】 喻凡；

【作者基本信息】 上海交通大学 ， 车辆工程， 2011， 博士

【摘要】 本文将非路面车辆和机器人学的相关技术相结合,对软地面智能气垫车自主导航问题进行了研究。气垫车能够利用垫升系统承担部分载荷,减小行走机构的接地压强,因此可以提高在沼泽、滩涂、沙漠等软地面环境下的通过性。辅以自主导航和自动控制技术,智能气垫车可自行开展诸如探险、搜救、清污、采样的危险性工作,保证了作业人员的安全。因此,本研究在石油业、矿业、运输业、农业等众多生产部门及军事领域具有应用价值。本文将智能气垫车自主导航体系抽象为气垫车、真实环境和虚拟环境三大系统,将自主导航的工作流程归纳为基于传感器信息的真实环境识别和虚拟环境构建,基于虚拟环境的车辆运动规划,以及用以实现规划目标的车辆运行控制。因此,全文依次研究了气垫车建模、土壤参数估值、能耗优化、运动规划和运行控制。首先,结合原理样车,建立了气垫车的结构模型和动力学模型。气垫车的主要结构包括垫升系统和驱动系统,针对垫升系统建立了其压强和流量的损失模型,针对驱动系统建立了其运动传递模型。此外,还利用拉格朗日法建立了11自由度车辆动力学模型。接下来,作为软地面环境识别的重要部分,对8个关键土壤参数进行了分步估值,依次利用g-EKF联合算法计算推力相关土壤参数,利用对数均值法计算垂向力相关土壤参数,利用最小二乘法计算推土阻力相关土壤参数。试验结果表明,分步估值算法能够消除估值多解问题,提高估值准确性。继而进行了能耗优化研究。分析了气垫车运行参数(如载荷分配比、滑转率和车速等)之间的关系,将总能耗简化表示为载荷分配比和滑转率的函数,并设计了自适应蜜蜂算法进行参数寻优。优化的结果可作为运行参数的稳态目标值,用于后续运动规划和运行控制。试验结果表明,由于自适应蜜蜂算法具有功能划分、并行计算和局部搜索范围自适应调节的特点,它能够同时提高优化准确性和优化效率。继而利用人工势场法对气垫车进行了运动规划,计算自主导航过程中其平动速度和转动速度的瞬态目标值。在建立势场时,多种运动要求以目标物吸引势、障碍物排斥势、动力学安全势和能耗经济势的形式予以表现。在运用势场时,对车速方向和速率进行分步规划,以提高计算效率。试验结果表明,在不同的工况下,各势函数项有不同的相对重要性,运动规划结果使当时主要的运动要求得以满足。最后研究了气垫车的运行控制。采用以模糊PID控制器为核心的并行双环控制系统控制车速和载荷分配比,模糊推理规则的设计体现出参数协调控制的思想。相对而言,车速控制跟踪瞬态目标值,因而更加注重动态性能;载荷分配比控制跟踪稳态目标值,因而更加注重静态性能。仿真试验结果表明,在控制系统的作用下,气垫车能够顺利抵达目的地并实现软着陆,其行驶路径和速度的变化符合理论分析结果;轮速控制表现出良好的动态性能,有利于提高车辆的行驶安全性;载荷分配比控制表现出良好的静态性能,有利于降低车辆能耗。本文虽以软地面智能气垫车作为研究对象,但其研究成果可向多个领域推广。例如,设计的土壤参数在线分步估值算法是一种针对气垫车辆的普适算法;自适应蜜蜂算法作为一种启发式全局优化算法,可用于一般的函数优化问题;改进的人工势场法在势函数构成、多种运动要求的协调以及运动规划的实施等方面均有所创新,可用于智能车辆和机器人的运动规划。上述三项技术也因此构成本文的创新点。更多还原

【Abstract】 By combining the techniques in the fields of off-road vehicle and robotics, this work studies the autonomous navigation problem for intelligent air-cushion vehicles (ACVs) for soft terrain. ACVs are able to use the lifting system to adjust the pressure under the travelling mechanism and thus able to improve the crossing ability on soft terrain, such as swamp, beach and desert. By taking advantage of autonomous navigation and automatic control techniques, intelligent ACVs may substitute people in dangerous tasks like exploration, searching, cleaning and sampling. Consequently, this interdisciplinary study could have practical significance in many applications in industry, agriculture and military.The autonomous navigation system of the targeted intelligent ACV is described with three sub-systems, namely, the ACV, a real environment, and a virtual environment. Its workflow is briefly summarized as the real environment identification and the virtual environment construction based on sensor information, vehicle motion planning implemented in the virtual environment, as well as vehicle operating control for the realization of the planning objectives. The structure of the work, correspondingly, comprises the ACV modeling, soil parameter estimation, energy consumption optimization, motion planning and operating control.Firstly, based on a prototype ACV, the structure model and dynamic model of the targeted ACV are established. In the perspective of structure, the ACV consists of a lifting system and a propulsion system. The loss models of air pressure and flow are built for the former and the motion transfer model for the latter. In the perspective of dynamics, an 11 degree-of-freedom system is modeled by the Lagrangian method.As an important aspect of identification for soft terrain environment, eight key soil parameters are estimated step-by-step. In succession, the hybrid g-EKF algorithm is employed for the calculation of the soil parameters related to the tractive effort, the logarithm-mean algorithm for those related to the vertical forces, and the least squares method for those related to the bulldozing resistances. Experiment results show that the estimation multi-solution problem can be eliminated and estimation accuracy can be improved.Energy consumption optimization is studied next. By analyzing the relationships among the ACV’s running parameters (e.g. load distribution, slip ratio and velocity), energy consumption is simplified as a function with only respect to slip ratio and load distribution ratio. Then the designed Adaptive Bees Algorithm (ABA) is applied to optimize this function. Its results, as the steady-state objectives of the running parameters, will be used in the succeeding motion planning and operating control. Experiment results show that the ABA can improve both optimization accuracy and optimization efficiency with the help of the features of functional partitioning, parallel operation and adaptive patch size adjustment.Motion planning is then made for the ACV by the Artificial Potential Field (APF) method, on the purpose of solving the instantaneous objectives of its linear velocity and angular velocity. For the APF modeling, multiple motion requirements are embodied in the contents of potential items, involving goal attractive potential, obstacle repulsive potential, dynamic safety potential and energy economy potential. In the APF application, the objectives of velocity direction and speed are determined respectively so as to improve calculation efficiency. Experiment results show that the potentials have different importance in applications and the major motion requirement could thus be satisfied.Finally, operating control is implemented to the ACV. A parallel double-loop control system is designed to control its linear/angular velocities and load distribution ratio. In the control system, fuzzy-PID controllers make up its core whose fuzzy reasoning rules present the idea of parameter coordinated control. Relatively, velocity control tracks an instantaneous objective, so that dynamic performance is paid more attention to. On the contrary, load distribution radio control tracks a steady-state objective and thus attaches more importance to static performance. Experiment results show that the ACV could, under control, reach the destination with soft-landing. The variations of path and velocity are consistent with the theoretical analysis in this process. Moreover, the velocity control presents a good dynamic performance, as required, to benefit the vehicle’s safety, and the load distribution ratio control manifests a good static performance to decrease energy consumption.Although a soft-terrain-used intelligent ACV is taken as the object of study here, the proposed techniques and research results could have broader applications. For example, the designed soil parameter estimation algorithm is universally reasonable for ACVs. The ABA, as a global optimization metaheuristic, is suitable for general functional optimization problems. Relying on the improvements to potential function composition, motion requirement coordination and motion planning implementation, the improved APF is worth taking up a place in the area of motion planning for intelligent vehicles and robots. Consequently, the above three respects constitute the innovations of the work.更多还原