Project : texmex
Section: Application Domains
Robotics and Visual Servoing
If collaboration between robotics and vision is an already old subject, it has undergone an important change of paradigm in the five last years. Hitherto, collaboration was considered on the level of planning: a camera observed the world around a robot to enable him to plan its displacements. The results appeared to be not so satisfactory.
The field of collaboration then moved towards control: the vision is not any more used to plan a movement, but to ensure its follow-up and good execution, by setting up a closed loop of control including vision   . The results are promising and many industrial applications already exist.
Some difficulties remain: the tasks to be achieved are specified using a target image that should be reached, but that assumes that the robot is able to establish a bond between this image and the current image provided by the camera. This is a classical image matching problem. If these two images do not have anything in common, it will be necessary to use a collection of intermediate images, which define intermediate positions of the robot before reaching the final position.
Therefore, the control problem corresponds to an image collection management problem, with dynamic collections to follow the evolution of the environment of the robot, and needs for fast access for recognition. This application appears important because it widely opens the experimental use conditions of visual servoing: once an environment collected in a base, the robot can start from any position to go towards any target. If this kind of approach presents little interest for articulated arm for which the articular co-ordinates can be read directly, an autonomous vehicle can benefit from it in restricted environments such as car parks. In this case, the systems of positioning as the GPS do not offer sufficient relative precision and do not give information of orientation.