allow lerobot-eval to work with kwargs

docs/source/envhub.mdx

@@ -284,6 +284,37 @@ envs_dict = make_env(
 
 Any keyword arguments you pass will be forwarded to the hub environment's `make_env` function. Check the environment's documentation for supported configuration options.
 
+### Using Custom kwargs with lerobot-eval
+
+When evaluating policies using the `lerobot-eval` CLI, you can pass custom kwargs to hub environments using the `--env_kwargs.` prefix:
+
+```bash
+lerobot-eval \
+    --policy.path=user123/example-policy-checkpoint \
+    --env=user123/example-sim-backend \
+    --eval.batch_size=1 \
+    --eval.n_episodes=10 \
+    --env_kwargs.task_id=demo_task_alpha \
+    --env_kwargs.agent_profile=arm_v2 \
+    --env_kwargs.target_item=object_red \
+    --env_kwargs.run_mode=offscreen \
+    --env_kwargs.enable_sensors=true \
+    --env_kwargs.record_output=true \
+    --env_kwargs.output_horizon=10 \
+    --env_kwargs.output_stride=15 \
+    --env_kwargs.state_features=joint_angles \
+    --env_kwargs.visual_streams=front_camera
+```
+
+All `--env_kwargs.*` arguments will be collected into a dictionary and passed as keyword arguments to the hub environment's `make_env` function. This allows you to:
+- Pass configuration file paths
+- Override default settings
+- Specify custom task parameters
+- Control simulation options (headless mode, camera settings, etc.)
+- Select different embodiments or objects
+
+The hub environment's `make_env` function receives these as regular keyword arguments, so check the environment's documentation for the available options.
+
 ## URL Format Reference
 
 The hub URL format supports several patterns:

src/lerobot/configs/eval.py

@@ -38,6 +38,8 @@ class EvalPipelineConfig:
     seed: int | None = 1000
     # Rename map for the observation to override the image and state keys
     rename_map: dict[str, str] = field(default_factory=dict)
+    # Additional kwargs to pass to hub environments (e.g., config_path, config_overrides, custom params)
+    env_kwargs: dict = field(default_factory=dict)
 
     def __post_init__(self) -> None:
         # HACK: We parse again the cli args here to get the pretrained path if there was one.

src/lerobot/scripts/lerobot_eval.py

@@ -43,6 +43,17 @@ lerobot-eval \
 
 Note that in both examples, the repo/folder should contain at least `config.json` and `model.safetensors` files.
 
+You can also evaluate a model on a Hub environment with custom kwargs:
+```
+lerobot-eval \
+    --policy.path=HF_USER/HF_REPO \
+    --env=HF_USER/HF_REPO \
+    --eval.batch_size=1 \
+    --eval.n_episodes=10 \
+    --env_kwargs.environment=env_A \
+    --env_kwargs.embodiment=emb_B \
+```
+
 You can learn about the CLI options for this script in the `EvalPipelineConfig` in lerobot/configs/eval.py
 """
 
@@ -509,7 +520,12 @@ def eval_main(cfg: EvalPipelineConfig):
     logging.info(colored("Output dir:", "yellow", attrs=["bold"]) + f" {cfg.output_dir}")
 
     logging.info("Making environment.")
-    envs = make_env(cfg.env, n_envs=cfg.eval.batch_size, use_async_envs=cfg.eval.use_async_envs)
+    envs = make_env(
+        cfg.env,
+        n_envs=cfg.eval.batch_size,
+        use_async_envs=cfg.eval.use_async_envs,
+        **cfg.env_kwargs,
+    )
 
     logging.info("Making policy.")