Image processing and performance

[j_johnson]

If you are using a class inherited from IterativeRobot, note that there is no built in delay, so it will use as much processing power as possible, the kernel splitting it equally with tasks of the same priority. To correct this, you can add a taskDelay() call in each of the XXXContinuous functions.

The delay should be long enough to cede as much processing power to other threads as possible without being so long that you are missing calls to the XXXPeriodic functions. (Assuming you are not using the XXXContinuous for anything time sensitive.) Make sure to include taskLib.h which declares this method.

Here is some documentation on VxWorks tasks if you are interested in reading more.

[nathanww]

You also probably want to take a look at your vision processing system and see if anything can be sped up there. Using the lowest resolution at which you can still identify the target will be helpful, as well as using a nested structure for target identification(for example, only look for a pink flag if a suitable green flag has already been found)

[gvarndell]

Originally Posted by j_johnson

If you are using a class inherited from IterativeRobot, note that there is no built in delay, so it will use as much processing power as possible, the kernel splitting it equally with tasks of the same priority.

Note that vxWorks will not, by default, do round-robin scheduling of tasks with the same priority. If there are 3 tasks with priority 100, and one of them never blocks (timed wait, request unavailable resource, etc.) then that task will forever be the only priority 100 task to run. Lower priority tasks will be locked out -- of course, higher priority tasks can still run.
To enable round-robin scheduling, the user code needs to call kernelTimeSlice() and pass it a tick count.
Those interested in knowing more about this stuff should use the Workbench HELP facility -- Help -> Help Contents.