How to obtain the position of the pig and position of the other agent? about malmo-challenge HOT 6 OPEN

Hi @luarsu,

The PigChaseSymbolicStateBuilder class gives you an example how to retrieve the positions of entities in the environment: https://github.com/Microsoft/malmo-challenge/blob/master/ai_challenge/pig_chase/environment.py#L58

Basically, when your retrieve a world_observation from the AgentHost, there is a field named 'entities' you can access which holds all the coordinates (x, y, z, yaw, name) for all the entities.

Let us know if you have more questions,
Hope it helps,

Morgan

from malmo-challenge.

Hi, @mfuntowicz
When I'm detecting the position of Agent_1 and target, sometimes there are errors. For example, they are not included in state information; or the position of Agent_1 doesn't change in 2 sequential checks.

from malmo-challenge.

Hi @Haishion,

Is this similar to issue #37 ?

from malmo-challenge.

@mfuntowicz yes, almost the same as #37
I lost target for a few times, and sometimes it's not in the play-area.

Besides, I just get a strange error as shown in following picture. This happened randomly after a number of episodes. I'm using your baseline file with my agent set as Agent_2.

from malmo-challenge.

Running the baselines "out of the box" will only run 500 steps. Does that match with the number of steps ran before you see this error ? In such case it's just that the experiment is done, but we don't gracefully detect it.

You can increase the number of epochs run on the baseline with the -e parameter (default to 5) in the command line:

python pig_chase_baseline.py -e 100 ...

Regarding your comment on the target outside the play-area, do you mean "outside the area delimited by the fences" ? In such case did it happen at the beggining of the experiment, or after some actions ?

The pig is randomly placed at each episode start. In the case you see this happen when a new episode is starting is something we definitively should take a closer look.

Morgan

from malmo-challenge.

Hi, @mfuntowicz
Thanks for your reply. However, as far as I can see, the error is not due to experiment getting done, since I've made the maximum number of loops large enough.

Yes, I mean that the target is sometimes out of the area delimited by the fences, and this can happen during an episode.

from malmo-challenge.