|
Zrehen,
Stephane Corridor Navigation Using Neural Fields This work describes an implementation of Amari's neural fields in the context of mobile robot control. The task is here to maintain a robot in the middle line of a richly textured unknown corridor, using the visual position of the vanishing point as a cue to the robot's attitude with respect to the middle line. As was shown by G. Schoner et al., neural fields attractor dynamics allow the robust visual control of a mobile robot, even in the presence of noise and unreliable data. The principle is to capitalize on time-accumulated evidence to take decisions for behavior control. In the present context, the rich visual structure of the corridor sometimes makes difficult the precise localization of the vanishing point. The work will describe the advantage of using neural fields for such a task.
|