Skip to content

orbitalgym.policies.action_repeat

Wrapper that calls an inner Policy every repeat env steps and re-emits its command in between. Pair it with POMDPAdapter(env, action_repeat=repeat) so the trajectory the env flies matches the one the planner searched. See How-to → Use the POMDP adapter.

Policy state

orbitalgym.policies.action_repeat.ActionRepeatState

Cached command, tick counter, and the inner policy's state.

step counts ticks since the wrapper was initialised. The inner policy runs when step % repeat == 0 — so on the first tick — and cached_dv holds the command replayed on every other tick.

Wrapper

orbitalgym.policies.action_repeat.ActionRepeatPolicy dataclass 

ActionRepeatPolicy(
    inner: Any,
    repeat: int,
    n_vehicles: int = 0,
    command_cls: Any = None,
)

Call inner every repeat env steps, replaying its command in between.

init_state 

init_state(
    config: Any = None,
    env_state: Any = None,
    key: Array | None = None,
    inner_init_state: Any = None,
) -> ActionRepeatState

Build the initial state with a zero cached command and the counter at 0.

The counter starts at 0, so the inner policy is called on the very first tick and the zero cache is never emitted. inner_init_state seeds the wrapped policy; when it is None the inner policy's own zero-argument init_state is used if it has one.