The Level 2 competition is the basis for most of the previous analysis material. In Level 2 the rovers must search the entire roving area for 10 samples and return them to the starting platform.
The analysis until now assumed the rovers would perform a systematic search of the roving area. To do this the location of each rover was needed with some precision to all the construction of a map of the area that was searched. The location of the rover when it collects a sample is also needed so the rover knows the direction to proceed to return to the starting platform.
The general analysis estimate that the rovers know their position to within 0.5 m and the bearing when driving straight to less that 1.0 degree. This may be difficult to achieve using only vision and it is uncertain that using an Inertial Measurement Unit (IMU) would be much better.
But is this really necessary? We will take a look.
The easier case is returning to the starting platform after collecting a sample. The exact location of the rover is not needed. All the rover needs is the general bearing to the starting platform. Vision processing can guide the rover once it is in sight of the white band shell and / or the starting platform. There should be a recovery capability if the rover unexpectedly encounters the boundary fence or other unexpected landmark.
Knowing the exact location of the rover when it collects a sample is not really critical.
What about while searching? The driver here is to avoid searching areas that are already covered. The assumption is to use a systematic search and map the locations covered.
Are there alternatives?
The easiest, perhaps, is to allow a random search.
The original Search Analysis was based on a single rover performing the search and retrieval. How does that analysis change if we use the three rovers used in 2013?
Previously one hour was allowed for searching and one hour for the collection and return of samples to the platform. With 3 rovers the search could be completed in (60 minutes / 3) 20 minutes and, similarly, the samples would be returned in 20 minutes. The challenge could be completed in 40 minutes which is really interesting.
Just to be safe, lets assume 30 minutes to return samples because I suspect the return is not as simple as dividing by 3 but do not want to dive into the details. That leaves 90 minutes for each rover to search for samples. (Or the equivalent of 270 minutes of search for a single rover.)
Since we know a rover can search 22,000 m2 in 60 minutes, the three rovers can search (270 min / 60 min * 22,000 m2) 99,000 m2 which covers the roving area 4.5 times. That much redundancy may be sufficient to allow a random search.
Time to revisit a search simulation I did for 2013 and modify the behaviors to allow covering already searched areas.