Cooperation

• Cooperative fire detection in Comets

  The first step of the mission scenario is fire detection. First, MARVIN follows a path while gathering sensor data with its fire sensor. Once an alarm (or several alarms) are detected, HELIV is sent to confirm (or discard) the alarm and localize it more precisely. Afterwards, in case of a confirmed alarm, a monitoring phase is initiated.

  Fire sensor as the "white nose" of MARVIN

  All the vehicles are able to estimate their positions by different means (DGPS and others) in a global coordinate frame. The position data label all the sensor data, and it will be used for localization of the alarms.

  The fire sensor data are processed to evolve a fire probability grid that covers the scenario. Initially, each cell of this map is set to a probability of 0.5; that is, nothing is known about the probability of having fire in a particular position. The cell size is about 5 meters, so that is the maximum resolution in positioning using the grid.

  Figure 1	Figure 2
  Figure 3	Figure 4
  Figure 5

  From the grid, connected regions with high probability are extracted, computing their mean position and their second order momenta (as a measure of the uncertainty of this mean position). These regions are declared as possible fire alarms.

  Then, the confirmation phase begins. During confirmation, Helivision infrared images are processed. Each time a new image arrives, the current alarms are projected over the image plane of the camera (using the positioning data that label the images). The alarms are associated with the detected objects on the image (or discarded if no fire was detected on the image in the position).

  Infrared Fusion

  Helivision

  The associated data are used to update the probability of being fire for each object, and also to refine its estimated position.

  In the example shown in the figures, three alarms are generated using the fire sensor data. Then, the confirmation phase is able to discard the two false alarms. Also, the position of the alarm is iteratively updated. Finally, the position of the alarm is determined with a precision close to one meter.

  	Easting	Northing	Height
  Measured position of the fire	564627	4443961	200
  Final estimated position (fusion)	564629	4443961	200.04
  Estimated standard deviation	1.5	2	0.28

  
  
  
• Movie on the distributed task allocation procedure

  This movie shows: Allocation of a few "goto" tasks. 4 robots and 50 "goto" tasks. Allocation of 2 constrained tasks (watch-out and communication relay). 4 robots and several various tasks.

• Illustration of the Plan Merging Operations (PMO) for distributed UAVs coordination

  PMO can be considered as an independent coordination process that should be performed continuously, in background and in real time, to ensure that UAVs trajectories are conflict free while they perform their respective plans. The space is considered as a ressource, and the spatial coordination of UAVs will be performed through a coherent, dynamic allocation of space (in terms of 3d cells or voxels), taking into account the expected time of occupation of the cells by the UAVs, along their respective trajectories.

  The figure 1 shows the trajectories of 2 UAVs (top view) which trajectories may be conflictual (red squares). UAVs are not yet aware of the risk. Their individual plans are depicted on figure 2.

  Figure 1

  Figure 2

  In order to ensure that their respective trajectories are conflict free (which is obviously currently not the case), UAV K and UAV M will perform a Plan Merging Operation of their respective plans of ressources occupation (the voxels of space). The figure 3 shows the application of PMO (actually, PMO is performed continuously). The insertion by UAV K of A4 and A5 in the merged plan is not possible : a conflict is raised, since these ressources are already planned by UAV M at a time which is the same time as UAV K was exepecting to occupy them.

  Figure 3

  The conflict can be solved in two ways:

  • if UAV K is able to slow down (significantly) its flying speed (case 1, on figure 4), then a delay will be introduced in its plan at the place where the occupation of voxels A4 and A5 were conflictual. Hence UAV K will preserve the structure of its initial plan, and will go on as soon as he will receive the confirmation signal from UAV M that voxels A4 and A5 are free.
  • if UAV K is not able to slow down enough then it will have to perform a re-planning (case 2, on figure 5), taking into account the current constraints of the merged plan (the occupation of A4 and A5 by UAV M). Hence UAV K will modify its trajectory accordingly.

  Figure 4

  Figure 5

• Experiments in May 2003

  In the videos below, Karma blimp, Marvin helicopter and the teleoperated helicopter are flying at the same time and the communications system was working well.

  Videos recorded during the experiments in May 2003.

  Each UAV received its mission from the Control Centre through the communications network.

  UAVs can cooperate to achieve tasks such as monitoring the fire. In the photograph, the autonomous helicopter (Marvin) and the teleoperated helicopter are taking and sending images of the same fire from different points of view.