Mission Performance for Autonomous Vehicles

Mission Performance for Autonomous Vehicles

The navigation system designed and implemented within the EMS-Vision framework consists of several components: the mission planning module (MP) is responsible for calculating optimal routes from any starting point to a goal point, either specified by the central decision unit (CD) or by a human operator (HO). For this purpose it uses various sources of background knowledge. On the one side, digital maps containing information about the operation area (roads, landmarks and impassable areas besides the roads) are used. On the other side, knowledge about the own kinematic limitations and behavioral capabilities implemented is stored. This covers basic availability as well as a statistical evaluation of the probability of success for the capabilities implemented. The content of the knowledge bases is discussed in detail in [Gre 98]. For determining the initial position, a standard GPS receiver is used. As at the moment mission performance is limited to road networks, planning is limited to the mapped road network. (Future work will concentrate on offroad scenarios),

Planning is always performed simultaneously with 3 different sets of optimization criteria, so that as planning result several routes to the desired goal are available. Essential information about these results, containing distance, estimated duration and the capabilities needed are provided via an interface to the decision modules. If one of these planning results is accepted by CD or HO, a mission plan is generated, otherwise replanning with varied criteria may be initiated. The mission plan contains a sequential list of mission elements specifying complex tasks for locomotion and references to objects relevant for perception and task performance, like (cross-) roads and landmarks expected along the route. During mission performance, the second navigation module, the Mission Expert (ME) is active. It continuously determines actual mission progress and compares it with the mission plan.

In order to improve actual position estimation, well recognizable natural 3D objects in the environment are used for localization. One major task of ME is the determination of points of transitions between elements of the mission plan. Therefore, several criteria are set for successful or failed transitions. The criteria may be generated by a combination of the following aspects:

  • Temporal aspects: duration of mission element or absolute point in time.
  • Spatial aspects: the stretch covered within one mission element or position relative to a static object.
  • Events: perception of an object expected, e.g. a crossroad.

The parameters for the criteria, e.g. the stretch covered, are dynamically calculated and set during mission performance by the Mission Expert considering static background knowledge and dynamic ego-state data. If a failure criterion is met, e.g. because an expected crossroad had not been detected, and thus it is not possible to fulfill a planned turn-off maneuver, this state is announced to CD. CD then may initiate a replanning. In order to achieve higher robustness and to have background knowledge for at least the time horizon necessary for replanning, alternative mission elements are inserted at node points within the road network, like at intersections.


Figure 16: Turn-off maneuver, images from the wide angle cameras

Download MPEG Video of turn-off maneuver (2min/18MB)

A detailed overview about mission performance may be found in [GrD 00].