Vision-based techniques for following paths with mobile robots. (Techniques de suivi chemin basées vision pour les robots mobiles)

In this thesis, we focus on the use of robot vision sensors to solve the path following problem, which in recent years, has been a popular target for engaged researchers around the world. Initially, we shall present the working assumptions that we used, along with the relevant kinematic and sensor models of interest, and with the main characteristics of visual servoing. Then, we shall attempt to survey the research carried out in the field of vision-based path following. Following this, we shall present two schemes that we designed for reaching and following paths. These respectively involve nonholonomic robots and legged robots. Both schemes require only some visible path features, along with a coarse camera model, and, under certain conditions, may guarantee convergence even when the initial error is large. The first control scheme is designed to enable nonholonomic mobile robots with a fi pinhole camera to reach and follow a continuous path on the ground. Two visual servoing controllers (position-based and image-based) have been designed. For both controllers, a Lyapunov-based stability analysis has been carried out. The performance of the two controllers is validated and compared by simulations and experiments on a car-like robot. The second control scheme is more application-oriented than the first and has been used in ASPICE, an assistive robotics project, to enable the autonomous navigation of a legged robot equipped with an actuated pinhole camera. The robot uses a position-based visual servoing controller to follow artificial paths on the ground, and travel to some required destinations. Apart from being a useful testing-ground for this path following scheme, the ASPICE project has also permitted the development of many other aspects in the field of assistive robotics. These shall be outlined at the end of the thesis.