fix: integrate PR #3315 review feedback

- ci: add Docker Hub login step, add HF_USER_TOKEN guard on eval step - docker: tie patches to pinned versions with removal guidance, remove unnecessary HF_TOKEN for public dataset, fix hadolint warnings - docs: fix paper link to arxiv, add teaser image, fix camera names (4→3 cameras), fix observation dims (480x640→240x320) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-05-25 05:29:55 +00:00 · 2026-04-16 18:33:53 +02:00
parent ad9662b4a8
commit a4102ee86d
4 changed files with 51 additions and 31 deletions
@@ -334,6 +334,12 @@ jobs:
        with:
          cache-binary: false

+      - name: Login to Docker Hub
+        uses: docker/login-action@v3 # zizmor: ignore[unpinned-uses]
+        with:
+          username: ${{ secrets.DOCKERHUB_LEROBOT_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_LEROBOT_PASSWORD }}
+
      # Build the full-install image: SAPIEN, mplib, CuRobo, pytorch3d +
      # simulation assets (~4 GB). Layer cache lives in the runner's local
      # Docker daemon — reused across re-runs on the same machine.
@@ -349,6 +355,7 @@ jobs:
          cache-to: type=local,dest=/tmp/.buildx-cache-robotwin,mode=max

      - name: Run RoboTwin 2.0 smoke eval (1 episode)
+        if: env.HF_USER_TOKEN != ''
        run: |
          # Named container (no --rm) so we can docker cp artifacts out.
          docker run --name robotwin-eval --gpus all \
@@ -60,8 +60,11 @@ RUN cd ${ROBOTWIN_ROOT}/envs \
       uv pip install -e . --no-build-isolation --no-cache

 # Upstream patches (mirror RoboTwin's script/_install.sh).
-# mplib: drop a broken `or collide` clause in planner.py.
-# sapien: fix URDF loader encoding + .srdf extension check.
+# These patches target the exact versions pinned above; re-check when upgrading.
+# mplib==0.2.1: drop a broken `or collide` clause in planner.py.
+#   Safe to remove once mplib > 0.2.1 ships with the fix upstream.
+# sapien==3.0.0b1: fix URDF loader encoding + .srdf extension check.
+#   Safe to remove once sapien > 3.0.0b1 ships with the fix upstream.
 RUN python - <<'EOF'
 import pathlib, re, site
 for d in site.getsitepackages():
@@ -80,7 +83,7 @@ EOF

 # Simulation assets from TianxingChen/RoboTwin2.0: embodiments (~220 MB) +
 # objects (~3.74 GB). background_texture (~11 GB) is intentionally skipped.
-ARG HF_TOKEN=""
+# The dataset is public — no auth token needed.
 RUN python - <<'EOF'
 import os, pathlib, zipfile
 from huggingface_hub import hf_hub_download
@@ -92,7 +95,6 @@ for fname in ("embodiments.zip", "objects.zip"):
        repo_id="TianxingChen/RoboTwin2.0",
        repo_type="dataset",
        filename=fname,
-        token=os.environ.get("HF_TOKEN") or None,
        local_dir=str(assets_dir),
    )
    with zipfile.ZipFile(local, "r") as z:
@@ -100,10 +102,14 @@ for fname in ("embodiments.zip", "objects.zip"):
    pathlib.Path(local).unlink()
 EOF

-RUN cd ${ROBOTWIN_ROOT} && python script/update_embodiment_config_path.py
+WORKDIR ${ROBOTWIN_ROOT}
+RUN python script/update_embodiment_config_path.py

 ENV PYTHONPATH="${ROBOTWIN_ROOT}:${PYTHONPATH}"

+# Return to the lerobot source directory (set by base image) before overlaying.
+WORKDIR /lerobot
+
 # Overlay the PR's source code on top of the nightly image.
 COPY --chown=user_lerobot:user_lerobot . .

@@ -2,22 +2,30 @@

 RoboTwin 2.0 is a **large-scale dual-arm manipulation benchmark** built on the SAPIEN physics engine. It provides a standardized evaluation protocol for bimanual robotic policies across 60 tasks with strong domain randomization (clutter, lighting, background, tabletop height, and language instructions).

- Paper: [RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation](https://robotwin-platform.github.io)
+- Paper: [RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation](https://arxiv.org/abs/2504.13495)
 - GitHub: [RoboTwin-Platform/RoboTwin](https://github.com/RoboTwin-Platform/RoboTwin)
 - Leaderboard: [robotwin-platform.github.io/leaderboard](https://robotwin-platform.github.io/leaderboard)
 - Dataset: [lerobot/robotwin_unified](https://huggingface.co/datasets/lerobot/robotwin_unified)

+<div style={{ textAlign: "center" }}>
+  <img
+    src="https://robotwin-platform.github.io/static/images/teaser.png"
+    alt="RoboTwin 2.0 benchmark overview showing dual-arm manipulation tasks"
+    style={{ maxWidth: "100%" }}
+  />
+</div>
+
 ## Overview

-| Property      | Value                                                      |
-| ------------- | ---------------------------------------------------------- |
-| Tasks         | 60 dual-arm manipulation tasks                             |
-| Robot         | Aloha-AgileX bimanual (14 DOF, 7 per arm)                  |
-| Action space  | 14-dim joint-space, continuous in `[-1, 1]`                |
-| Cameras       | `head_camera`, `front_camera`, `left_wrist`, `right_wrist` |
-| Simulator     | SAPIEN (not MuJoCo)                                        |
-| Eval protocol | 100 episodes/task, 50 demo_clean demonstrations            |
-| Eval settings | **Easy** (`demo_clean`) and **Hard** (`demo_randomized`)   |
+| Property      | Value                                                    |
+| ------------- | -------------------------------------------------------- |
+| Tasks         | 60 dual-arm manipulation tasks                           |
+| Robot         | Aloha-AgileX bimanual (14 DOF, 7 per arm)                |
+| Action space  | 14-dim joint-space, continuous in `[-1, 1]`              |
+| Cameras       | `head_camera`, `left_camera`, `right_camera`             |
+| Simulator     | SAPIEN (not MuJoCo)                                      |
+| Eval protocol | 100 episodes/task, 50 demo_clean demonstrations          |
+| Eval settings | **Easy** (`demo_clean`) and **Hard** (`demo_randomized`) |

 ## Available tasks

@@ -160,14 +168,13 @@ lerobot-eval \

 ## Camera configuration

-By default, all four cameras are included:
+By default, all three cameras are included:

 | Camera key     | Description                    |
 | -------------- | ------------------------------ |
-| `head_camera`  | Overhead / third-person view   |
-| `front_camera` | Front-facing static camera     |
-| `left_wrist`   | Left arm wrist-mounted camera  |
-| `right_wrist`  | Right arm wrist-mounted camera |
+| `head_camera`  | Torso-mounted overhead view    |
+| `left_camera`  | Left arm wrist-mounted camera  |
+| `right_camera` | Right arm wrist-mounted camera |

 To use a subset of cameras, override `--env.camera_names`:

@@ -176,7 +183,7 @@ lerobot-eval \
  --policy.path="your-hf-policy-id" \
  --env.type=robotwin \
  --env.task=beat_block_hammer \
-  --env.camera_names="head_camera,left_wrist,right_wrist" \
+  --env.camera_names="head_camera,left_camera" \
  --eval.batch_size=1 \
  --eval.n_episodes=10
 ```
@@ -185,15 +192,15 @@ lerobot-eval \

 Key parameters for `RoboTwinEnvConfig`:

-| Parameter            | Default                                             | Description                        |
-| -------------------- | --------------------------------------------------- | ---------------------------------- |
-| `task`               | `"beat_block_hammer"`                               | Comma-separated task name(s)       |
-| `fps`                | `25`                                                | Simulation FPS                     |
-| `episode_length`     | `300`                                               | Max steps per episode              |
-| `obs_type`           | `"pixels_agent_pos"`                                | `"pixels"` or `"pixels_agent_pos"` |
-| `camera_names`       | `"head_camera,front_camera,left_wrist,right_wrist"` | Comma-separated active cameras     |
-| `observation_height` | `480`                                               | Camera pixel height                |
-| `observation_width`  | `640`                                               | Camera pixel width                 |
+| Parameter            | Default                                  | Description                        |
+| -------------------- | ---------------------------------------- | ---------------------------------- |
+| `task`               | `"beat_block_hammer"`                    | Comma-separated task name(s)       |
+| `fps`                | `25`                                     | Simulation FPS                     |
+| `episode_length`     | `300`                                    | Max steps per episode              |
+| `obs_type`           | `"pixels_agent_pos"`                     | `"pixels"` or `"pixels_agent_pos"` |
+| `camera_names`       | `"head_camera,left_camera,right_camera"` | Comma-separated active cameras     |
+| `observation_height` | `240`                                    | Camera pixel height                |
+| `observation_width`  | `320`                                    | Camera pixel width                 |

 ## Leaderboard submission

@@ -583,7 +583,7 @@ class RoboTwinEnvConfig(EnvConfig):

    RoboTwin 2.0 is a dual-arm manipulation benchmark with 60 tasks built on the
    SAPIEN simulator. The robot is an Aloha-AgileX bimanual platform with 14 DOF
-    (7 per arm). All four cameras are enabled by default.
+    (7 per arm). All three cameras are enabled by default.

    See: https://robotwin-platform.github.io
    Dataset: https://huggingface.co/datasets/lerobot/robotwin_unified