Define / list use-cases of editor calling Env

I suggest the following cases for the Editor calling the Env (including transition map etc). Ideally these should be agreed with the developers of the Core so that the interaction can be kept simple and not too brittle. (Maybe introduce a Facade for the editor to call rather than operating directly on internals like Env, Rail, Grid, NN Agent + Env agents, etc)

• Editor adds new transition

  • env adds reverse transition (or should Editor do this?)

• Editor removes a transition

  • env or editor removes the reverse transition?

• Editor adds new agent initial location (and initial direction chosen by env?)

• Editor adds new agent target location.

• Editor changes agent initial location, target loc, initial direction.

• Editor adds new live agent, either at initial location, or elsewhere.

• Editor calls step() to move all agents on.

  • currently editor calls play_model Player.step() which
    • calls act on the NN agent to populate agent actions
    • calls step() on env, saves observation ready for next Agent.act()

• Editor calls Save()

• Regenerate Rail

  • Editor creates a new env, with a random rail generator, no of agents, new obs builder?

• Resize Rail - do we support resize (shrink + grow) existing rail? Or just start afresh with a new env?

The use cases above are largely implemented now.

Please review the current version of the editor. It should be done.

mentioned in commit f51c77cb

mentioned in merge request !68 (merged)

mentioned in commit 2f23359a

closed

Gone. Closed.

changed milestone to %v1