module moplayground.eval.rollout

---

function rollout_policy

rollout_policy(
    env,
    config,
    tradeoff=None,
    T=10.0,
    camera='track',
    width=1080,
    height=720
)

Roll out a trained policy on env and return rendered frames.

Loads the policy specified by config (single-objective if config.mo2so.enabled else multi-objective with the given tradeoff), JIT-compiles the inference function, and delegates the actual stepping/rendering to minimal_mjx.eval.rollout_policy.

Args:

• env: The environment to roll out in (typically built with create_environment).
• config: Run config (ConfigDict) for the trained policy.
• tradeoff: Per-objective weighting used to condition the multi-objective policy. Defaults to all-ones with length equal to the number of objectives in config. Ignored when config.mo2so.enabled.
• T: Rollout duration in seconds.
• camera: MuJoCo camera name to render from.
• width: Frame width in pixels.
• height: Frame height in pixels.

Returns: Whatever minimal_mjx.eval.rollout_policy returns — at the time of writing a 4-tuple (frames, reward_plotter, ...).