Skip to content

Observations

Choose an observation mode with the observation argument to make_env():

from mudgym import make_env

env = make_env(observation="parsed")
observation, info = env.reset()
env.close()

Every observation includes text and points, including custom field selections. The base environment tracks points from the initial quickscore and subsequent game events, and uses that same score to calculate rewards.

text

Contains the common text and points observations only.

from mudgym import make_env

env = make_env(observation="text")
observation, info = env.reset()
env.close()

print(observation["text"])
Beaten track.
You're on a rough east-west track with a dense forest to the north and pasture to the south. It is raining.

parsed

Adds keyed data parsed from game output.

Key Value
points 200
room_name beaten track near cliff
room_name_index 38
here cliff, road
features cliff, road
mobiles empty
players empty
vitals [59, 59, 54, 54, 67, 67, 0, 59]
flags [0, 0, 0, 0]
reset_minutes 105
weather fair
weather_index 1
available_exits [1, 1, 1, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 1]
available_exit_names north, east, south, northeast, southeast, southwest, out, swampward
portables empty
inventory empty
Beaten track near cliff. You are at the end of a rough track. There is a dangerous cliff to the west marked "Lovers' Leap".

available_exit_names lists the directions available according to the exits command.

vitals is [stamina, max_stamina, effective_strength, strength, effective_dexterity, dexterity, magic, max_magic], and flags is [blind, deaf, crippled, dumb].

cheats

Adds hidden state output from the mgcheats command, some of which wouldn't typically be known to a player. Most notably room_id.

Key Value
points 200
vitals [60, 60, 56, 56, 64, 64, 0, 60]
flags [0, 0, 0, 0]
reset_minutes 105
weather fair
weather_index 1
available_exits [1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 0, 1]
available_exit_names north, east, south, west, northeast, southeast, southwest, northwest, out, swampward
room_id mtrack2
room_id_index 688
room_name beaten track
room_name_index 37
fighting 0
dark 0
glowing 0
asleep 0
gifted 0
here road
players empty
mobiles empty
portables empty
inventory empty
Beaten track. You're on a rough east-west track with a dense forest to the north and pasture to the south.

bytes

Adds raw_bytes, a uint8 NumPy array padded or truncated to 16,384 bytes. The full, unpadded bytes are available as info["raw_bytes"].

raw = info["raw_bytes"]

Use RawBytesField(max_bytes=N) to change the array size.

Shown as a bytes literal here for readability:

b"\x1b[32mBadly-paved road\x1b[37m.\n\x1b[0;32;40mYou are standing on a badly-paved road, which runs from the east to stop at a large wall constructed to the west. There is a narrow gap in this blockage, but it is so tight that if you wanted to go that way you'd have to drop everything to get through. North and south are the footh"
...
Badly-paved road. You are standing on a badly-paved road, which runs from the east to stop at a large wall constructed to the west. There is a narrow gap in this blockage, but it is so tight that if you wanted to go that way you'd have to drop everything to get through. North and south are the foothills of a pair of majestic mountains, and northeast is a deep valley. It is raining. A streetsign has fallen here.

Creating your own keys

Custom observations are defined by selecting fields and the keys each field contributes:

from mudgym import make_env
from mudgym.envs.fields import FEScoreField, SuperQuickLookField

env = make_env(
    field_parsers=(
        SuperQuickLookField(include_keys=("room_name", "here")),
        FEScoreField(include_keys=("vitals",)),
    )
)
env.close()

You can add your own field parsers in the same way by creating an ObservationField subclass. Take a look at the fields in mudgym/envs/fields/ to use as a reference.