NotAnyMike/HRL

Ideas to improve Recovery_delayed/direct

Closed this issue · 1 comments

  • Give much less time to recover in recovery delayed
  • Use the discount rate (or think if it would be useful)
  • Add bias towards acceleration of zero
  • Check if outside track
  • Punish behaviour in which goes to the track really fast, get in the track and continue going fast so it leaves the track again
  • Punish behaviour in which the recovery ends with sliping (due to very fast acceleration)

There is no point in keeping these two apart, merging into only one module