Threshold Criteria
Percent Improvement of Resolution
The resolution of the previous robot positions are compared with the newest positions. If the improvement in resolution realized by the movement of the robots is lower than 5%, then the robots are told to not move. This is done because resolution improvements of small increments indicate the robots are close to optimally placed. When the resolution is improved by more than 5%, the controller continues to update the robot positions in order to optimize the array’s parameters. This was done to prevent the robots from stopping when the resolution is too large for accurate estimations. The percentage does not need to be 5%; it can be adjusted from the main control interface. A higher threshold causes the robots to move fewer times but also causes them to stop with less accurate estimation of the source. Lower threshold values yield more accurate measurements but require the robots to move more times and take longer to optimize the array parameters.
Reactivating After Source Movement
Reactivating when source moves: The threshold improvement criterion worked well for fixed, nonmoving sound sources. However, when the sound source was moved, the robots did not optimize to new positions. The source movement caused a decline in the resolution of estimation, because the source was naturally being moved away from the region of highest resolution. As described above, this resolution would be compared to the previous resolution to determine whether or not to initiate movement. This comparison always confirmed that the pre-movement resolution was higher, so movement was never initiated. This challenge prevented the system from adaptively responding to source movement.
A new criterion was implemented to fix the issue. This criterion needed to be capable of determining whether movement of the source had occurred. When movement did, in fact, occur the comparator between current and previous resolution would be ignored and the algorithm would be forced to attempt to find a new position. An issue with this was that the source estimate naturally changes as the robots move because the web of possible points shifts and changes with every new robot position. Thus, having the controlling algorithm react to any change in the source estimate at all would lead to a never ending attempt by the robots to continue trying to converge to an optimal location. To prevent this, we computed the distance of estimated source movement and compared this to our critical resolution variable, which represents the average distance from the current estimation to its neighbors in the web of possible points. By doing this we could get a sense of whether the source estimate was jumping around in its given neighborhood or whether a significant movement of the source had occurred.
If the source change was less than this critical resolution value, then the movement was ignored as being within the margin of error described by the critical resolution and the robots then responded as determined by the percentage improvement criterion. If the movement was found to be far greater than the resolution, then the source change can only be explained by a movement of the source. As described above the algorithm then forced the system to optimize the robot positioning for this new source location.
