Mohammadi Arash, Asadi Houshyar, Mohamed Shady, Nelson Kyle, Nahavandi Saeid
IEEE Trans Cybern. 2019 Sep;49(9):3471-3481. doi: 10.1109/TCYB.2018.2845661. Epub 2018 Jun 26.
Driving simulators are effective tools for training, virtual prototyping, and safety assessment which can minimize the cost and maximize road safety. Despite the aim of a realistic motion generation for the impression of real-world driving, motion simulators are bound in a limited workspace. Motion cueing algorithms (MCAs) aim to plan an acceptable motion feeling for drivers, without infringing the simulated boundaries. Recently, model predictive control (MPC) has been widely used in MCAs; however, the tuning process for finding the best weights of the MPC optimization is still a challenge. As there are several objectives for the optimization without any standard weighting for solution evaluations, a nonbiased scalarization of solutions for the purpose of comparison is impossible. In this paper, a clear method for obtaining the best MPC weighting has been proposed. This method searches for the best tune of MPC cost function weights, reduces the user burden for weight tuning while receiving feedback from the user satisfaction. The MPC-based MCA weights are optimized using a multiobjective genetic algorithm (GA) considering objectives, such as minimization of motion inputs (linear acceleration and angular velocity), input rates, output displacements and the sensed motion errors. Any process based on trial-and-error has been omitted. The adjusted weights have to satisfy a set of predefined conditions related to maximum tolerated error and maximum displacement. The obtained Pareto-front is used for decision making via an interactive GA (IGA), aiming for maximization of the decision maker's satisfaction. A Web interface is developed to interact with the IGA and to influence the region of searching. Simulation results show the superiority of the proposed method compared with the previous empirical tuning method. The sensed motion error is minimized using the proposed method and with the same available workspace, a more realistic motion can be rendered to the driver.
驾驶模拟器是用于培训、虚拟原型制作和安全评估的有效工具，可将成本降至最低并最大限度地提高道路安全性。尽管旨在生成逼真的运动以营造真实驾驶的感觉，但运动模拟器受限于有限的工作空间。运动提示算法（MCA）旨在为驾驶员规划可接受的运动感觉，同时不超出模拟边界。最近，模型预测控制（MPC）已广泛应用于MCA；然而，寻找MPC优化最佳权重的调整过程仍然是一个挑战。由于优化有多个目标，且没有用于解决方案评估的标准权重，因此无法对解决方案进行无偏标量化以进行比较。本文提出了一种获得最佳MPC权重的清晰方法。该方法搜索MPC成本函数权重的最佳调整，在接收用户满意度反馈的同时减轻用户的权重调整负担。基于MPC的MCA权重使用多目标遗传算法（GA）进行优化，考虑的目标包括运动输入（线性加速度和角速度）最小化、输入速率、输出位移和感知运动误差。省略了任何基于试错的过程。调整后的权重必须满足一组与最大容忍误差和最大位移相关的预定义条件。通过交互式GA（IGA）使用获得的帕累托前沿进行决策，旨在最大化决策者的满意度。开发了一个Web界面与IGA进行交互并影响搜索区域。仿真结果表明，与先前的经验调整方法相比，该方法具有优越性。使用该方法可将感知运动误差降至最低，并且在相同的可用工作空间下，可以为驾驶员呈现更逼真的运动。
