search

mo3pheus / MarsRover

A mars rover simulation project that will serve as a test bed for machine learning algorithms.
GNU General Public License v3.0
15 stars 10 forks source link

SCLK and PositionSensor get out of sync #5

Open mo3pheus opened 6 years ago

mo3pheus commented 6 years ago

Problem: The SpacecraftClock and PositionSensor for the Rover can get out of sync ( a minute apart or more ).

Cause: The SpacecraftClock and PositionSensors run on separate threads and get kicked off at different times, Over a period of time due to thread synchronization and waits for lock acquisition, they can get out of sync. If the timeScaleFactor is 60 - these discrepancies can add up quickly.

Solution: There is a command you can issue to the Rover for clockSynchronization - this syncs up the spaceCraftClock and the positionSensor. Unfortunately right now the rover waits for this command to be explicitly issued from the commandCenter.

Ideally the rover would have a tune up thread that measures the difference between the spaceCraftClock and the positionSensor Clock and triggers a sync if that difference gets above a certain threshold.

Solution II - tried implementing this but it failed due to thread sync issues. But ideally you could have the PositionSensor as an observer to the spaceCraftClock - i.e. when the spaceCraftClock updates, so does the positionSensor. This makes a cleaner solution than having the positionSensor ticking on a separate thread and dependent upon periodic forced syncs. The clock Syncs are not without their cost - the pacemaker and roverGarbageCollector are both restarted when this happens. Also conceptually the spaceCraftClock should be the final word upon what time it is on the spaceCraft. This is the computer's internal clock and all components rely upon it for their time. Look to how the TelemetrySensor is tied to the trackingAnimationEngine to see how this could be done.

Things to learn from this issue: 1) Thread synchronization. 2) Usage of the marsConfig properties file in configuring the rover. 3) Understanding of the timeScale factor that affects all rover communications. 4) If SolutionII is selected, the observer design pattern.

mo3pheus commented 6 years ago

Matt - not to pressure you into doing something you don't want to - but if you decide you want an issue to look at - this could be a good start.

mo3pheus commented 6 years ago

roverDevel is where all bugFixes go to.

mo3pheus commented 6 years ago

2017-12-17 20:42:01,610 [pool-4-thread-1] INFO space.exploration.mars.rover.kernel.SpacecraftClock - Internal time::2016-01-01T03:03:00.000-06:00 sclk of::1/01210:06:41:14:91150 2017-12-17 20:42:07,691 [pool-4-thread-1] INFO space.exploration.mars.rover.kernel.PositionSensor - Houston - this is curiosityActual.ephemerisTime: 5.04889748183912E8 stateCuriosity: -2.21331994261E8 stateCuriosity: -1.12446652913E8 stateCuriosity: -4.1574799206E7 stateCuriosity: 28.484 stateCuriosity: -15.194 stateCuriosity: -7.162 owltMSLEarth: 839.632 stateEarth: 2.2136176861E8 stateEarth: 1.12461782101E8 stateEarth: 4.158039376E7 stateEarth: -28.492 stateEarth: 15.194 stateEarth: 7.163 owltEarthMSL: 839.745 positionEarthWRTCuriosity: 2.2136176861E8 positionEarthWRTCuriosity: 1.12461782101E8 positionEarthWRTCuriosity: 4.158039376E7 owltEarthMSL2: 839.745 angSepHGAEarth: 112.029 utcTime: "2016-01-01~03:08:00" sol: 1210 sclkValue: " 1/01210:06:46:06:88479"

mo3pheus commented 6 years ago

for Sol 1210 - in a few seconds the computer clock and the positionSensor clock are 5 minutes apart. With a timeScaleFactor of 60, every second for us means 1 minute for the rover.

This was done so that we wouldn't have to wait 5 minutes for the messages to go back and forth.