Gymnasium Compatibility

Supported Gymnasium spaces and value conversion behavior.

RLMesh follows the Gymnasium environment shape:

import gymnasium as gym
from rlmesh import EnvFactory


class CartPole(EnvFactory):
    def make(self):
        return gym.make("CartPole-v1")


CartPole().serve("127.0.0.1:5555")

The server reads observation_space, action_space, reset, step, and close. Common Gymnasium wrappers can stay in place.

Spaces

RLMesh supports these Gymnasium spaces. Stable spaces follow the compatibility guarantees in Compatibility, while Experimental spaces may still change.

Fundamental Spaces

Gymnasium space	RLMesh space	Stability	Notes
`Box`	`Box`	Stable	Uniform and array bounds are accepted.
`Discrete`	`Discrete`	Stable	`start` is preserved.
`MultiBinary`	`MultiBinary`	Experimental	Integer and shaped forms are accepted.
`MultiDiscrete`	`MultiDiscrete`	Experimental	One- and two-dimensional `nvec` are supported.
`Text`	`Text`	Experimental	Custom charsets are preserved; default charsets are treated as unbounded.

For Text, RLMesh preserves min_length and max_length. Only Gymnasium’s default alphanumeric charset is treated as unrestricted, so punctuation and whitespace are accepted when the source space uses that default.

Composite Spaces

Gymnasium space	RLMesh space	Stability	Notes
`Tuple`	`Tuple`	Experimental	Supported when child spaces are.
`Dict`	`Dict`	Stable	Supported when child spaces are.

Not supported:

Graph
Sequence
OneOf

Unsupported spaces fail directly instead of silently changing the environment contract.

Value Conformance

RLMesh checks values against their declared spaces a little differently from Gymnasium’s passive_env_checker:

An out-of-bounds Box value, or an out-of-charset/length Text value, is delivered with a warning (under the rlmesh.conformance.warning info key) rather than passed through silently, and NaN is always rejected. See Compatibility for the strict/off policy knob.
A value is coerced to its declared dtype before transport, whereas Gymnasium warns but forwards the original dtype. A float64 observation for a float32 space is delivered as float32.

Conversion Helpers

Use these helpers when you need to convert spaces yourself:

from rlmesh import spaces

rlmesh_space = spaces.from_gymnasium_space(gym_space)
gym_space = spaces.to_gymnasium_space(rlmesh_space)