Skip to Main content Skip to Navigation

Commande des plateformes avancées de simulation de conduite

Abstract : Driving simulators are advanced devices composed of four components: a virtual scene projected on a wide screen to imitate the road and the traffic, an audio system to play the driving sounds (horn, squeal of brakes, etc.), a car cockpit (including a real dashboard, the pedals and the seat of the driver) to copy the body position and the interaction of the driver with a real vehicle and finally a robot carrying the car cockpit to provide its motion. While the first three components could be considered as offering a sufficiently high degree of realism, the robot presents a very low capacity of displacement, thus preventing if from performing the real car motions. In fact, the aim of a driving simulator is not tracking real trajectories produced by outdoors driving but reproducing the corresponding motion sensations. How could we then, generate realistic motion sensations in simulation despite the constrained robot motion? It is the aim of Motion Cueing Algorithms (MCA) to give heuristically an answer to this problem. We thought however that, if we know the maximum capacity of restituting motion sensations} then we will be able to answer a better question: how could we generate, using the simulator, the ''best'' motion sensations despite the constrained robot motion? And then using this knowledge, we would be able to calibrate existing MCAs in order to maximize their performances or we could even design in an optimal way the geometry and the dynamics of the simulation robot. In this thesis, we offer an answer to this question thanks to our: Maximum Performance Algorithm (MPA). The MPA covers all aspects linked to the simulation. It integrates the notion of motion sensation through perception models taken from the literature. It includes, as well, our proper detailed robot model to account for its nonlinear dynamics. And it ensures the respect of all physical limits in the optimization process by using a nonlinear programming approach. The complexity of this problem is not only due to the difficulty of defining a measure of performance (what is the exact definition of sensation?) but also to the constrained robot motion (multiple level of constraints: position, speed and acceleration) and to its nonlinear dynamics. Once all the required notions were set, our approach was tested on a simulator with and without the motion of the base. The results were rich in information: the high and low frequency characterization of the hexapod translation and the tilt coordination respectively, the non causality of the algorithm, the very limited robot capacity and the necessity of using a scaling factor. Two variations of the MPA were then introduced to deal with the last points. Two performance indexes were introduced to measure the quality of simulation in terms of magnitude and profile tracking. In summary, this thesis presents a set of tools that are very useful to study the simulator behavior on typical scenarios and to calibrate the robot or the control algorithm. Two other points where addressed in this dissertation: the redundancy problem and the robust robot control. As for the first point, the MPA shows that, in redundant robots, the rails and the hexapod present overlapping bandwidths in the high frequency domain. So how could we benefit from this redundancy? The exploitation of this capability is currently done by frequency separation methods without taking into account this frequency overlapping. Within this bandwidth, these two degrees of freedom could be considered as equivalent. Our aim is to use this equivalence to improve the motion restitution. We offer two algorithms based on the hybrid systems framework which deal with the longitudinal mode. Their goal is to improve the restitution of motion sensations by reducing false cues (generated by actuators braking) and decreasing null cues (due to actuators blocking). As for the second point, it deals with the tracking control of robot systems in presence of perturbations such as modeling errors and disturbance forces. More specifically, this paper aims at reviewing the well-established Robust Computed Torque (RCT) controller. Designing a RCT scheme consists in both, selecting subclasses of robot models and, establishing conditions on the control parameters leading to the robustness of the system. Generally, it amounts to the elaboration of a gain threshold beyond which robustness is achieved. One challenging problem is to develop the minimum threshold for the less conservative conditions on the control and the model. We had two contributions: The Encompassing Formalization (EF) and a refinement of a former result. The EF is an extension of the RCT formalization based on the Lyapunov direct method developed by Qu and Dawson. Then for the specific RCT scheme (developed by Samson), EF combined with passivity property will be used to elaborate lower gain thresholds. This result is presented as a theorem for which an original proof is proposed.
Document type :
Domain :
Complete list of metadatas

Cited literature [87 references]  Display  Hide  Download
Contributor : Ecole Mines Paristech <>
Submitted on : Friday, May 4, 2007 - 8:00:00 AM
Last modification on : Thursday, September 24, 2020 - 5:22:33 PM
Long-term archiving on: : Thursday, September 30, 2010 - 7:42:59 PM


  • HAL Id : pastel-00002213, version 1


Hatem Elloumi. Commande des plateformes avancées de simulation de conduite. domain_other. École Nationale Supérieure des Mines de Paris, 2006. Français. ⟨pastel-00002213⟩



Record views


Files downloads