rlmesh.torch.SandboxVectorEnv

classExperimentalfrom rlmesh.torch import SandboxVectorEnv[source]

Experimental Torch-backed owned sandbox session for vectorized environments.

SandboxVectorEnv(
    source: str,
    /,
    num_envs: int,
    *,
    vectorization_mode: str | None = None,
    build: SandboxBuild | None = None,
    runtime: SandboxRuntime | None = None,
    connect_timeout_seconds: float = SANDBOX_REMOTE_CONNECT_TIMEOUT_SECONDS,
    **params: object,
) -> None

The sandbox starts multiple isolated environment instances and exposes them through the same vector client interface as a separately served endpoint.

Parameters

Name	Type	Description
`source`	`str`	A gym id / `gym://` / `hf://` source built from source, or a prebuilt rlmesh-serving image (`docker://img` / bare `img:tag`).
`num_envs`	`int`	Number of environment instances to create.
`vectorization_mode`	`str \| None`	Vectorization mode requested inside the sandbox.
`build`	`SandboxBuild \| None`	Optional `SandboxBuild` -- build-from-source infrastructure; ignored for a prebuilt image.
`runtime`	`SandboxRuntime \| None`	Optional `SandboxRuntime` -- `docker run` settings (`gpus` / `devices` / `volumes`); prebuilt-image source only.
`**params`	`object`	Environment construction params -- the binding forwarded to the factory's `make` (validated in the container before construction).

Examples

>>> from rlmesh.torch import SandboxVectorEnv
>>> envs = SandboxVectorEnv("CartPole-v1", num_envs=2)
>>> observations, infos = envs.reset(seed=42)
>>> envs.close()

Attributes

sandbox
sourceOriginal sandbox source string requested by the caller.
num_envsNumber of environment instances served by the endpoint.
single_observation_spaceObservation space for one environment in the vector.
single_action_spaceAction space for one environment in the vector.
observation_spaceAlias for single_observation_space.
action_spaceAlias for single_action_space.
addressEndpoint address this client is connected to.
env_idThis connection's container id (UUIDv7).
selected_workflow_editionThe workflow edition this connection negotiated with the env.
env_contractEnvironment contract returned by the endpoint handshake.
render_modeConfigured render mode reported by the endpoint.
metadataEndpoint metadata reported by the environment contract.
observation_space_specNative observation space spec reported by the endpoint.
action_space_specNative action space spec reported by the endpoint.

Methods

close()Detach the remote client and stop the owned sandbox container.
reset()Reset all remote environments and decode the observations.
step()Step all remote environments with a batch of actions.
render()Render a frame from the remote environment.
shutdown()Request owner-level shutdown of the remote environment endpoint.

sandbox

attribute[source]

sandbox

source

property[source]

property source: str

Original sandbox source string requested by the caller.

close

function[source]

close() -> None

Detach the remote client and stop the owned sandbox container.

num_envs

property[source]

property num_envs: int

Number of environment instances served by the endpoint.

single_observation_space

property[source]

property single_observation_space: Space[ValueT]

Observation space for one environment in the vector.

single_action_space

property[source]

property single_action_space: Space[ActionT]

Action space for one environment in the vector.

reset

function[source]

reset(
    *,
    seed: int | list[int] | None = None,
    options: dict[str, object] | None = None,
) -> tuple[ValueT, ResetInfo]

Reset all remote environments and decode the observations.

Parameters

Name	Type	Description
`seed`	`int \| list[int] \| None`	Optional seed or per-environment seed list.
`options`	`dict[str, object] \| None`	Optional reset options forwarded to the vector environment.

Returns

Type	Description
`ValueT`	Batched decoded observations and reset info dictionary. Besides the
`ResetInfo`	env's own keys, the info carries the runtime-reserved
`tuple[ValueT, ResetInfo]`	`episode_ids` (the runtime-minted episode id per lane). An env key
`tuple[ValueT, ResetInfo]`	of the same name wins and shadows it; the client logs that once.

step

function[source]

step(
    actions: ActionT,
) -> tuple[ValueT, ValueT, ValueT, ValueT, StepInfo]

Step all remote environments with a batch of actions.

Parameters

Name	Type	Description
`actions`	`ActionT`	Batched actions accepted by the vector endpoint.

Returns

Type	Description
`ValueT`	Batched observations, rewards, terminations, truncations, and info.
`ValueT`	Besides the env's own keys, the info carries two runtime-reserved
`ValueT`	keys on every step: `episode_ids` (the runtime-minted episode id
`ValueT`	per lane) and `completed_episodes` (how many lanes' episodes
`StepInfo`	ended on this step, `0` when none did). An env key of the same
`tuple[ValueT, ValueT, ValueT, ValueT, StepInfo]`	name wins and shadows the runtime's; the client logs that once.

address

property[source]

property address: str

Endpoint address this client is connected to.

env_id

property[source]

property env_id: str

This connection’s container id (UUIDv7).

A stable correlation identity, distinct from the human env name (env_contract.id).

selected_workflow_edition

property[source]

property selected_workflow_edition: str

The workflow edition this connection negotiated with the env.

The highest edition both sides can drive, at or below whatever either declared, in the exact spelling they agreed on – so a prerelease build reports its cohort, not the bare sealed base. Named apart from the workflow_edition declaration surfaces on purpose: this is what was agreed, not what this side asked for.

env_contract

property[source]

property env_contract: EnvContract

Environment contract returned by the endpoint handshake.

render_mode

property[source]

property render_mode: str | None

Configured render mode reported by the endpoint.

metadata

property[source]

property metadata: Metadata

Endpoint metadata reported by the environment contract.

observation_space_spec

property[source]

property observation_space_spec: SpaceSpec

Native observation space spec reported by the endpoint.

action_space_spec

property[source]

property action_space_spec: SpaceSpec

Native action space spec reported by the endpoint.

render

function[source]

render(*, env_index: int = 0) -> ValueT | None

Render a frame from the remote environment.

Parameters

Name	Type	Description
`env_index`	`int`	Environment index to render. Single environments use `0`.

Returns

Type	Description
`ValueT \| None`	A decoded render frame, or `None` when the environment has no frame.

shutdown

function[source]

shutdown(reason: str = 'owner shutdown') -> bool

Request owner-level shutdown of the remote environment endpoint.