RL Training track

[steviet91]

@Thonners - Thoughts on the training track:

• Add an is_open flag to the Track object to prevent trying to close the loop
• Start Finish point will be at the start. StartFinish line will be at the end
• Tracks will only be able to achieve an NLapsComplete of 1, i .e. need to set done when we hit it
• Do we want a random frequency wave generator to make the track?
• Do we want it to be a set of tracks? i.e for each episode a different track is loading, thus preventing the agents from getting fixated on the first feature it sees? If we keep the tracks short memory shouldn't be a problem, we could just load from the track set at run time. Include a set track method in the vehicle model to update this (and the vis).

[Thonners]

On the set of tracks, they could become progressively more difficult as they are completed - no point giving it an impossible task when it's failed 100 times on an easy one, but if it's learnt the simple 'go' then we could look at adding features to train it a bit at a time.

Probably worth implementing it as 'each time it completes a track, a harder one is added to the track list, and for every episode, a track is selected at random from the list.' That should help stop it from forgetting how to solve the easy tracks.

[Thonners]

I'll push my rlagent branch today too, so you can hopefully merge what you've done on the OpenAI gym formatting so we're working on the same page/code