UMR CNRS 7253

Considering the various studies on the current use of maps for intelligent vehicles, we made the observation that there is still a real demand in this area and current maps are not well adapted to this purpose. That's why we have oriented this action towards the definition and specification of maps for intelligent vehicles. This work is based on the knowledge of maps done for mobile robotics (large spatial coverage, less details, symbolized objects and semantics for robot navigation in controlled environment) and maps done for GIS (small spatial coverage, lots of details, low level representations for human navigation / understanding).

Questions:

• What information should IV maps record / provide?
• How to generate, update and represent IV maps?
• Which format? Need of standardization?
• How to make use of IV maps?
• Etc.

Given the complexity and variety of scenes that make up the environment of a vehicle and given evolutions at different time scales, we agreed to look the map at the street level. We then defined two main components: the static and dynamic part. The static part consists of several levels corresponding to different uses.

Partners : All (IGN)

The objective of this work is to study the solutions to generate and update the various attributes of maps for intelligent vehicles and this at all static description levels. The following paragraph presents certain issues addressed in this action. Most geographic databases are generated with the help of aerial and satellite images, which can cover relatively large area, but fail to capture details at the street level because of the limited spatial resolution and point of view. Today, maps in car navigation systems do not contain details such as road marks, road signs or zebra zones which are essential for intelligent vehicles in their task of driving assistance, motion planning and decision making. Mobile mapping systems (MMS) (or Intelligent Vehicle Systems (IVS)) can be considered as complementary tools for mapping, extracting the road details from a different point of view.

Partners : SJTU, IGN, AIT

The objective of this action is to study possible solutions to generate dynamic components of the maps for intelligent vehicles and at different levels of description. From a methodological point of view, the approaches are very different depending on the time scale and techniques of data collection (intelligent vehicles, sensors associated with the infrastructure). Observation methods, data fusion, data analysis and machine learning methods will be developed by the in this action.

Partners CSIS, PKU, LIAMA, HEUDIASYC

• Perception and localization, Heudiasyc, SJTU, IGN ==
• Situation understanding, risk evaluation: E-motion, PKU, Liama ==

These two case studies were defined according to the original proposal. They will be described in more detail in the next progress report. Several discussions already took place regarding the use of maps available in each country. In France, public research or higher education may use maps of the IGN Bati3D (layer RGE). In contrast, Chinese partners can not access an equivalent. The maps available via the Internet are much less accurate and there is no possibility of purchasing ones. However, solutions like OpenStreetMap (maps constructed from user inputs via the Internet) have been discussed and could be considered