Closed vrodriguezf closed 8 months ago
It does not affect the observation space: roll vs no roll observations.xlsx
I ran many trials and found a few that matched up the Mission_Elapsed_Time (between trials, the time is often off by a few game ticks, which are each 0.02s, so I had to run multiple before getting some pairs that match).
The files show the observations and the pairwise differences. Between trials, they are often off by 1 tick, but other than that, there is no difference Ex: between [no roll #3] and [no roll #4], we see the evader is off by 13m and 41m, which is how far it travels in 0.02s, or one game tick. Note that this occurs regardless of if we roll or not. So this difference is not relevant. Other than this noisy difference, there is no real difference.
but I will look into the updated action space reference frame option to see if this is still relevant. I think it lets us use x/y/z reference frame, which might be easier (then we don't need the orientation)
Where does the noise come from? Why is it not a zero different exactly if the initial conditions of the game are the same?
I believe its because KSP might start simulating the evader and/or pursuer at different game ticks when loading in. I calculated and the velocity times 0.02s (which is the time of one tick) gives you the 13 and 41 difference. It might also be that there is some lag between getting the evader and pursuer's positions in lincoln lab's python code, which makes it off by 1 tick
CLosing this
Please let's double check empirically if rolling the spacecraft affects any part of the observation space.