INSAIT-Institute/LADDER-RobotArena-Long

LADDER RobotArena-∞-Long

47 WidowX tasks of 1 to 10 subgoals in RobotArena-∞ scenes, on Genesis. It is one of the three suites of LADDER, a long-horizon manipulation benchmark, packaged as a LeRobot EnvHub environment. Each simulator needs its own Python environment, so each suite is its own repository:

Suite	Repository	Simulator	Robot	Tasks
RoboLab-Long	INSAIT-Institute/LADDER-RoboLab-Long	Isaac Sim 6	Franka (DROID)	44
SimplerEnv-Long	INSAIT-Institute/LADDER-SimplerEnv-Long	SAPIEN 3 / ManiSkill 3	WidowX	54
RobotArena-∞-Long	this one	Genesis	WidowX	47

Evaluate a policy

From a LeRobot checkout, with Docker and the NVIDIA Container Toolkit:

pip install -e ".[dataset,evaluation,remote]"  # plus your policy's extra
lerobot-env-server --envhub=INSAIT-Institute/LADDER-RobotArena-Long@v1.0.0 --docker=true
lerobot-eval --policy.path=<your WidowX checkpoint> \
  --env.type=envhub --env.hub_path=INSAIT-Institute/LADDER-RobotArena-Long@v1.0.0 \
  --env.server.endpoint=tcp/127.0.0.1:7448 --env.task=scene54_split_2 \
  --eval.batch_size=8 --eval.n_episodes=8 --eval.use_async_envs=false

• The first --docker=true builds the simulator image (about 16 GB, 10 to 20 minutes); later starts reuse it.
• --env.task takes a task, a comma-separated list, or nothing for the whole suite (task list). A task's 8 episodes run as one batch, so keep the batch size at 8. The server builds the suite's five scenes before it reports ready (about 3 minutes the first time).
• To run it in-process instead, pip install "lerobot[dataset,evaluation]>=0.7.0" -r requirements.txt and replace --env.server.endpoint with --env.trust_remote_code=true.

Interface

Observation	pixels.main: 480x640 RGB; agent_pos: end-effector pose [x, y, z, rx, ry, rz] (base frame, axis-angle, BridgeData convention) and gripper (1 open)
Action	[dx, dy, dz, drx, dry, drz, gripper]: translation in the base frame, rotation in the end-effector frame, applied to the commanded pose as LADDER runs its Bridge policy; gripper 1 open, 0 closed
Control rate	5 Hz
Score	pc_success: LADDER's task success. avg_sum_reward: its subgoal progress at the end of the episode (also info["metrics"]["subgoal_progress"])

Relation to LADDER

ladder/ is LADDER's code at release v1.0.0, unchanged, run with the settings of LADDER's own server. The simulation assets (240 MB) are downloaded from that release on first use. Its reset frames match LADDER's recorded episodes, and LADDER's Bridge policy run through it succeeds as often as in LADDER's published episodes (21 and 20 of 32 against 21 on four tasks). LADDER's own harness stays the reference for the leaderboard.

Citation

@article{ladder2026,
  title   = {{LADDER}: A Long-Horizon Robot Manipulation Benchmark},
  author  = {Petkov, Petko and Albanese, Giuliano and Nikolov, Nikolay and Van Gool, Luc and Tremblay, Jonathan and Paudel, Danda Pani},
  journal = {arXiv preprint arXiv:2610.18053},
  year    = {2026},
  url     = {https://arxiv.org/abs/2610.18053}
}

License

MIT, as LADDER. Downloaded assets keep their own licenses (THIRD_PARTY_NOTICES.md).