Sim Environments

Scene YAMLs under scenes/ follow a three-tier hierarchy, DeployScene ⊆ SimScene ⊆ BenchmarkScene. Each tier has its own directory, its own loader-strictness gate, and its own CLI consumer. The conceptual overview, decision matrix, authoring guide, and per-backend scene.id catalogue all live in the in-tree scenes/README.md; this page is the per-file catalogue, one row per YAML.

Scene dependencies are auto-installed on first use (auto-install is on by default; set OPENRAL_AUTO_INSTALL_DEPS=0 to be prompted instead, e.g. when not in CI). Most scenes need an opt-in dependency group first: just sync --group <name> (sim / libero / robocasa / metaworld / maniskill3), never a bare uv sync.

RoboCasa is a special case: just sync --group robocasa provides only robosuite + deps, while the RoboCasa fork itself is git-cloned + installed editable at runtime by the HAL (ensure_backend_deps('robocasa_kitchen')). LIBERO and RoboCasa pin conflicting robosuite versions, so swap groups per task. Full recipe → Managing the Python environment & dependency groups.

An install step's output streams to your terminal, and its tail is quoted in the ROSConfigError if the step fails: when a backend or sidecar won't provision, read the quoted lines, not the exit code. Inside a ROS node the raw output is scattered through the launch log and the reasoner reports only goal_rejected, so the quoted tail is the only place the cause travels.

A sidecar pinned to wheels that don't exist for your platform (usually aarch64) shows up there as a uv "no source distribution or wheel for the current platform" line naming the package; see aarch64 CUDA hosts for the per-sidecar status on GB10 / DGX Spark and Jetson Thor.

Quick CLI

# DeployScene — env-only playground (reasoner picks the rSkill at runtime).
openral deploy sim --config scenes/deploy/openarm_tabletop.yaml

# SimScene — single rollout; supply the policy at the CLI.
openral sim run --config scenes/sim/libero_spatial.yaml --rskill smolvla-libero

# BenchmarkScene — paper-comparable single-scene eval; writes
# rskills/<vla>/eval/<scene_id>.json with reproduced_locally=true.
openral benchmark scene --config scenes/benchmark/libero_spatial.yaml \
                        --rskill smolvla-libero

# Benchmark suite — multi-scene aggregate (lives in benchmarks/, not scenes/).
openral benchmark run --suite libero_spatial --rskill smolvla-libero

Override flags (--task, --instruction, --max-steps, --n-episodes, --robot for free-axis scenes) work on every tier except benchmark run, which intentionally rejects them to guarantee suite reproducibility. See scenes/README.md for the full override matrix.

DeployScene catalogue (`scenes/deploy/`)

Env-only "robot + scene" pins. No task: block, no eval; the runtime reasoner picks the rSkill. Consumed by openral deploy sim, or by openral deploy run for the real-hardware scenes.

Config	Fixed / declared robot	`scene.id`	Backend	Use
`libero_pnp.yaml`	`franka_panda` (scene-fixed)	`libero_spatial`	LIBERO (robosuite + MuJoCo)	Boot LIBERO in deploy mode so a reasoner can issue arbitrary pick-and-place commands
`openarm_tabletop.yaml`	`openarm` (scene-fixed)	`openarm_tabletop_pnp`	Custom MJCF	OpenArm bimanual tabletop sandbox; default top camera matches the mddoai dataset POV
`robocasa_pnp.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceCounterToCabinet`	RoboCasa (MuJoCo)	Mobile-base kitchen pick-and-place sandbox; reasoner-driven
`behavior_r1pro.yaml`	`r1pro` (scene-fixed)	`behavior`	BEHAVIOR-1K / OmniGibson (Isaac Sim sidecar)	Full deploy graph on public `turning_on_radio` instance 0
`so101_box.yaml`	`so101_follower` (scene-fixed)	`so101_box`	Custom MJCF	100×61.5×75 cm box arena + OAK-D Pro overhead + wrist camera; deploy sandbox
`so101_bench.yaml`	`so101_follower` (scene-fixed)	`so101_bench`	Custom MJCF	SO-101 bench-arena deploy sandbox
`libero_object.yaml`	`franka_panda` (scene-fixed)	`libero_object`	LIBERO (robosuite + MuJoCo)	Boot LIBERO-Object in deploy mode; reasoner-driven pick-and-place
`robocasa_baguette.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceCounterToCabinet`	RoboCasa (MuJoCo)	Baguette pick-and-place kitchen sandbox; declares its ADR-0097 place target (`sim:cab_1_left_group_main`, read off the rounds 5/6 artifacts and confirmed live at seed 1 — region measured, allowance published). Carries the canonical recipe for reading a place target off a live MuJoCo body table
`robocasa_sink_cup.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceCounterToSink`	RoboCasa (MuJoCo)	Cup counter→sink pick-and-place; generalization sibling of `robocasa_baguette`. Declares its ADR-0097 place target (`sim:sink_island_group_main`, body 319 — corrected against a live seed-1 run after the HAL refused #142's guess fail-closed)
`robocasa_drawer_utensil.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceCounterToDrawer`	RoboCasa (MuJoCo)	Utensil counter→drawer pick-and-place; reach-down place target. No ADR-0097 declaration: its target is one of many generated `stack` levels, so a guessed name would resolve to a different real receptacle instead of failing closed. Confirmed live at seed 1 — the only bodies matching `drawer` are the fridge's freezer drawers, while the counter drawer is `stack_2_left_group_3_*`
`robocasa_fridge_drawer.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceFridgeShelfToDrawer`	RoboCasa (MuJoCo)	Vegetable fridge-shelf→fridge-drawer pick-and-place; enclosed workspace. Declares its ADR-0097 place target (`sim:fridge_main_group_fridge_drawer0`, body 98 — corrected against a live seed-1 run after the HAL refused #142's guess fail-closed; the drawer, not the fridge root, whose subtree would also exempt both doors)
`robocasa_navigate.yaml`	`panda_mobile` (scene-fixed)	`robocasa/NavigateKitchen`	RoboCasa (MuJoCo)	Mobile-base kitchen navigation sandbox (Nav2 graph compatible)
`robocasa_deliver_straw.yaml`	`panda_mobile` (scene-fixed)	`robocasa/DeliverStraw`	RoboCasa (MuJoCo)	The carry-while-navigating scene for issue #108: the base drives 3.795 m at the pinned seed 3 while holding the straw, which is what exercises `openral_nav2_bringup` mid-carry (the scan payload filter; the dynamic payload footprint was removed in PR #186). In `target50`, so XR-1 stays in distribution. Drawer is closed at reset — open → grasp → carry → place. See the carry survey
`robocasa_vslam.yaml`	`panda_mobile_vslam` (declared)	`robocasa/PickPlaceCounterToCabinet`	RoboCasa (MuJoCo)	Lidar-less twin of `robocasa_pnp`: the base localises from two base-mounted RGB cameras via in-process PyCuVSLAM, not a `LaserScan`
`robocasa_vslam_mono.yaml`	`panda_mobile_vslam` (declared)	`robocasa/PickPlaceCounterToCabinet`	RoboCasa (MuJoCo)	Single-camera variant: one RGB camera + DepthAnything-3 metric depth into cuVSLAM RGB-D odometry and nvblox
`g1_vln.yaml`	`g1` (declared)	`g1_vln_bench`	Bare MuJoCo twin	Unitree G1 VLN bench: head camera, ADR-0089 pinned walking controller, `BODY_TWIST` navigation skills
`openarm_zed_octomap.yaml`	`openarm` (declared)	`openarm_tabletop_pnp`	Custom MJCF HAL + real ZED-M	Real ZED depth through the octomap leg into the world map, with the simulated OpenArm HAL
`openarm_bench.yaml`	`openarm` (declared)	`openarm_bench`	Real hardware (`deploy run`)	OpenArm v2 real bench cell: real arms and real cameras
`openarm_real_world_voxels.yaml`	`openarm` (declared)	`openarm_real_world_voxels`	Real hardware (`deploy run`)	Real cell with the safety kernel's world-voxel check on; follow the runbook
`openarm_real_autonomous.yaml`	`openarm` (declared)	`openarm_real_autonomous`	Real hardware (`deploy run`)	`openarm_real_world_voxels` plus the reasoner, open-vocab detector and spatial-memory ingest: an LLM chooses the motion
`galaxea_a1_bench.yaml`	`galaxea_a1` (declared)	`galaxea_a1_bench`	Real hardware (`deploy run`)	Galaxea A1 deployment; start the ROS 1 sidecar, camera bridge, LingBot policy server and A1 runtime policy gateway first
`isaac_franka_urdf.yaml`	`franka_panda` (scene-fixed)	`isaac_sim`	Isaac Sim	Franka sandbox loaded from URDF (Isaac Sim)
`isaac_panda_mobile_urdf.yaml`	`panda_mobile` (scene-fixed)	`isaac_sim`	Isaac Sim	Mobile-base Panda loaded from URDF (Isaac Sim)
`isaac_openarm_warehouse.yaml`	`openarm` (declared)	`isaac_sim`	Isaac Sim	Bimanual OpenArm imported from its URDF, standing at a pallet of three pickable YCB props in the `Simple_Warehouse`; its meshes come from Enactic's public `openarm_description`, fetched on first use
`isaac_panda_mobile_warehouse.yaml`	`panda_mobile` (declared)	`isaac_sim`	Isaac Sim	Mobile-base Panda spawned in an aisle of NVIDIA's `Simple_Warehouse` (three shelf racks, lidar/SLAM/Nav2), three pickable YCB props on a pallet 5 m down the aisle. The generic template: `scene.assets_uri` (environment USD) + `robot_id` + `base_pose` (spawn) — see `scenes/README.md`

Deploy scenes are timing-faithful: one scene step is one control period of the robot (action_spec.control_freq_hz), and SimAttachedHAL.connect refuses a scene whose tick is anything else (openral_core.check_time_base, 1 %).

Isaac derives its physics substeps at boot (non-integer ratios refused); see Isaac Sim: any stage, any robot.
RoboCasa and LIBERO deploy scenes pin backend_options.control_freq_hz to the robot's rate (robosuite's physics time step is shrunk to divide the period); without the pin both keep robosuite's native 20 Hz, the benchmark contract the eval tier and every published LIBERO number run at.
Bare MuJoCo twins (MujocoArmHAL: OpenArm, SO-100/101, UR, Franka, ...) derive their physics steps per action from the same rate at connect (issue #358).

Isaac camera images trail the physics by 2–4 steps (the RTX pipeline), so each observation carries image_time_ns and deploy sim stamps /openral/cameras/* with the time each image shows, not the publish time (issue #361).

Every Isaac step also carries info["robot_contacts"] (that step's PhysX robot-vs-world contacts), and openral sim run folds them into each episode's contact_stats in summary.json: which links touched what, for how many steps, how deep (max_penetration_m) and how hard (max_impulse_ns). That is the evidence for which contacts a successful episode needs, to compare against the safety kernel's stops.

Isaac safety-kernel stops carry ground truth too: each sim.estop_ground_truth_snapshot line (backend: "isaacsim") lists the PhysX robot-vs-world contact pairs of the stop's physics step (link, other prim, min_separation_m (negative = penetration), contact count, impulse). PhysX has no shape-to-shape distance, so unlike MuJoCo there are no near-miss probes beyond the contact-offset band (contacts_caveat).

SimScene catalogue (`scenes/sim/`)

DeployScene + a single task: block. One CLI invocation, one or more EpisodeResults; sized for ad-hoc development and smoke tests. The policy comes from --rskill <name>; scene YAMLs do not pin a VLA. Consumed by openral sim run.

Isaac SimScenes can grade placement prompts: backend_options.placement_success (IsaacPlacementSuccess) maps a prompt, the scene's task.instruction verbatim, to the scene objects it asks to place and named world-frame regions (target and neighbours, e.g. shelf slots). Once an object rests for settle_s (neither moving nor turning), it is graded by the scene's grades rubric (IsaacPlacementGrade: name, score, target / adjacent region, whether it must be upright; first match wins, else fail 0). The default rubric is complete (in its target, upright: 1.0), partial (in its target, fallen: 0.5) and semi (in a neighbouring region: 0.25); a scene declares its own levels to grade differently. The episode takes its worst object's grade. Reaching the first grade terminates the episode with success; otherwise the episode's total_reward is its final grade's score, and every step's info["placement_grade"] reports the grade so far. Without the block an Isaac scene has no task reward.

Under openral deploy sim the scene is task-less (<scene>/_hal_deploy_noop, empty instruction), so no rubric is active at boot and every prompt's rubric ships to the sidecar. The HAL follows /openral/reward/active_task, where the rSkill runner publishes each ExecuteRskill goal's prompt on accept and "" at its end: the goal's prompt, verbatim, selects tasks[prompt] (sidecar set_task); a prompt with no entry is not graded (one sim.placement_grade_unavailable line). Each grade change logs sim.placement_grade and the goal's end sim.placement_grade_final (closing and best grade); the rubric keeps grading the settle after the goal ends, until the next prompt replaces it. A deploy run never terminates on the grade -- it is only reported (see telemetry). openral sim run still resolves the scene's own task.instruction at build and refuses one without an entry.

task:
  instruction: "put the cracker box on the shelf"
scene:
  backend_options:
    placement_success:
      regions:
        shelf_1: {min_xyz: [-0.70, 4.75, 0.20], max_xyz: [-0.40, 4.95, 0.50]}
        shelf_2: {min_xyz: [-0.70, 4.95, 0.20], max_xyz: [-0.40, 5.25, 0.50]}
      tasks:
        "put the cracker box on the shelf":
          - {object: cracker_box, target: shelf_1, adjacent: [shelf_2]}

Config	Fixed / declared robot	`scene.id`	`task.id`	Notes
`libero_spatial.yaml`	`franka_panda` (scene-fixed)	`libero_spatial`	`libero_spatial/0`	LIBERO-Spatial smoke; ad-hoc sibling of `scenes/benchmark/libero_spatial.yaml`
`openarm_tabletop.yaml`	`openarm` (scene-fixed)	`openarm_tabletop_pnp`	`openarm/pnp_cube_to_drawer`	Bimanual cube-to-drawer; mirrors the mddoai dataset POV
`robocasa_gr1_pnp_cup_to_drawer.yaml`	`gr1` (scene-fixed)	`robocasa/gr1/PnPCupToDrawerClose`	`robocasa/gr1/PnPCupToDrawerClose/0`	RoboCasa GR1 humanoid tabletop pnp
`robocasa_panda_mobile_kitchen.yaml`	`panda_mobile` (scene-fixed)	`robocasa/NavigateKitchen`	`robocasa/NavigateKitchen/0`	Mobile-base kitchen navigation; `deploy sim` Nav2 graph compatible
`robocasa_pnp.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceCounterToCabinet`	`robocasa/PickPlaceCounterToCabinet/0`	RoboCasa kitchen pnp smoke
`so101_tube_insertion.yaml`	`so101_follower` (scene-fixed)	`so101_box`	`so101_box/tube_insertion`	Box-arena tube-insertion smoke; geometry/sensors/spawn ranges configurable via `BoxSceneOptions`
`tabletop_cube_push.yaml`	`so101_follower` (free-axis default; pass `--robot` to override)	`tabletop_push`	`tabletop_push/push_to_goal`	Robot-agnostic cube push-to-goal
`widowx_carrot_on_plate.yaml`	`widowx` (scene-fixed)	`simpler_env`	`simpler_env/widowx_carrot_on_plate`	SimScene sibling of the SimplerEnv WidowX carrot benchmark; used by the OpenVLA-OFT issue #55 reproduction path
`aloha_transfer_cube.yaml`	`aloha_bimanual` (scene-fixed)	`aloha_transfer_cube`	`aloha_transfer_cube/0`	gym-aloha bimanual cube-transfer smoke
`franka_tabletop_push.yaml`	`franka_panda` (scene-fixed)	`tabletop_push`	`tabletop_push/push_to_goal`	Franka variant of the robot-agnostic cube push-to-goal
`pusht.yaml`	`pusht_2d` (scene-fixed; 2-D pymunk)	`pusht`	`pusht/0`	gym-pusht 2-D push smoke
`behavior_turning_on_radio.yaml`	`r1pro` (scene-fixed)	`behavior`	`behavior/turning_on_radio`	Official public-test instance 0 through the OmniGibson evaluator sidecar
`xr1_robocasa_pnp.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceCounterToCabinet`	`robocasa/PickPlaceCounterToCabinet/0`	XR-1 observation/action integration check; not Xiaomi's RoboCasa v0.2 benchmark
`xr1_robocasa365_open_drawer.yaml`	`panda_mobile` (scene-fixed)	`robocasa/OpenDrawer`	`robocasa/OpenDrawer/0`	200-step XR-1 RoboCasa365 review rollout, not the target50 evaluation
`xr1_robocasa365_close_blender_lid.yaml`	`panda_mobile` (scene-fixed)	`robocasa/CloseBlenderLid`	`robocasa/CloseBlenderLid/0`	150-step XR-1 RoboCasa365 approval rollout, not the target50 evaluation
`xr1_vlabench_select_fruit.yaml`	`franka_panda` (scene-fixed)	`vlabench`	`vlabench/select_fruit`	Short XR-1 wiring check (front/base/wrist camera order, NF4 rSkill); not a benchmark score

BenchmarkScene catalogue (`scenes/benchmark/`)

SimScene + required metadata: BenchmarkMetadata (paper URL + honest_scope) + non-None seed and n_episodes. The shipped values match the canonical paper protocol; openral benchmark scene against one of these writes rskills/<vla>/eval/<scene_id>.json with reproduced_locally=true. Most are also aggregated into a multi-scene suite under benchmarks/ (below).

Config	Fixed / declared robot	`scene.id`	`task.id`	`n_episodes`	Paper
`aloha_insertion.yaml`	`aloha_bimanual` (scene-fixed)	`aloha_insertion`	`aloha_insertion/0`	200	ALOHA / ACT
`aloha_transfer_cube.yaml`	`aloha_bimanual` (scene-fixed)	`aloha_transfer_cube`	`aloha_transfer_cube/0`	200	ALOHA / ACT
`libero_spatial.yaml`	`franka_panda` (scene-fixed)	`libero_spatial`	`libero_spatial/0`	500	LIBERO
`maniskill_pick_cube.yaml`	`franka_panda` (scene-fixed)	`maniskill3`	`maniskill3/PickCube-v1`	500	ManiSkill3
`metaworld_push.yaml`	`sawyer` (scene-fixed)	`metaworld`	`metaworld/push-v3`	50	MetaWorld MT10/MT50
`metaworld_pick_place.yaml`	`sawyer` (scene-fixed)	`metaworld`	`metaworld/pick-place-v3`	50	MetaWorld MT10/MT50
`metaworld_button_press.yaml`	`sawyer` (scene-fixed)	`metaworld`	`metaworld/button-press-v3`	50	MetaWorld MT10/MT50
`behavior_turning_on_radio.yaml`	`r1pro` (scene-fixed)	`behavior`	`behavior/turning_on_radio`	1	BEHAVIOR-1K
`metaworld_door_open.yaml`	`sawyer` (scene-fixed)	`metaworld`	`metaworld/door-open-v3`	50	MetaWorld MT10/MT50
`metaworld_drawer_open.yaml`	`sawyer` (scene-fixed)	`metaworld`	`metaworld/drawer-open-v3`	50	MetaWorld MT10/MT50
`pusht.yaml`	`pusht_2d` (scene-fixed; 2-D pymunk)	`pusht`	`pusht/0`	200	Diffusion Policy
`rlbench_open_drawer.yaml`	`franka_panda` (scene-fixed)	`rlbench`	`rlbench/open_drawer`	25	RLBench / 3D Diffuser Actor
`rlbench_meat_off_grill.yaml`	`franka_panda` (scene-fixed)	`rlbench`	`rlbench/meat_off_grill`	25	RLBench / 3D Diffuser Actor
`rlbench_close_jar.yaml`	`franka_panda` (scene-fixed)	`rlbench`	`rlbench/close_jar`	25	RLBench / 3D Diffuser Actor
`widowx_carrot_on_plate.yaml`	`widowx` (scene-fixed)	`simpler_env`	`simpler_env/widowx_carrot_on_plate`	24	SimplerEnv
`libero_object.yaml`	`franka_panda` (scene-fixed)	`libero_object`	`libero_object/0`	50	LIBERO
`libero_goal.yaml`	`franka_panda` (scene-fixed)	`libero_goal`	`libero_goal/0`	50	LIBERO
`libero_10.yaml`	`franka_panda` (scene-fixed)	`libero_10`	`libero_10/0`	50	LIBERO (= LIBERO-Long)
`robocasa_pnp.yaml`	`panda_mobile` (scene-fixed)	`robocasa/PickPlaceCounterToCabinet`	`robocasa/PickPlaceCounterToCabinet/0`	50	RoboCasa
`robotwin_beat_block_hammer.yaml`	`aloha_agilex` (scene-fixed)	`robotwin`	`robotwin/beat_block_hammer`	100	RoboTwin 2.0
`robotwin_handover_block.yaml`	`aloha_agilex` (scene-fixed)	`robotwin`	`robotwin/handover_block`	100	RoboTwin 2.0
`robotwin_lift_pot.yaml`	`aloha_agilex` (scene-fixed)	`robotwin`	`robotwin/lift_pot`	100	RoboTwin 2.0
`robotwin_place_empty_cup.yaml`	`aloha_agilex` (scene-fixed)	`robotwin`	`robotwin/place_empty_cup`	100	RoboTwin 2.0
`robotwin_stack_blocks_two.yaml`	`aloha_agilex` (scene-fixed)	`robotwin`	`robotwin/stack_blocks_two`	100	RoboTwin 2.0
`vlabench_select_fruit.yaml`	`franka_panda` (scene-fixed)	`vlabench`	`vlabench/select_fruit`	50	VLABench

n_episodes and seed ship at the paper-canonical value. Overriding --n-episodes on openral benchmark scene is allowed (useful for cheap smoke runs that don't claim paper-reproduction); the resulting RSkillEvalResult records the lowered count.

Multi-scene aggregations (e.g. all 10 LIBERO-Spatial tasks, benchmarks/metaworld_mt10.yaml (10 tasks) and benchmarks/metaworld_mt50.yaml (50 tasks), all 4 SimplerEnv WidowX tasks) live in benchmarks/. A suite YAML is a bare list[BenchmarkScene] at the YAML root; suite-level invariants (uniform robot_id, seed, n_episodes, and full metadata block) are enforced by openral_core.raise_on_invalid_suite.

Justfile shortcuts

The repo's Justfile groups sim-* recipes by which CLI they drive:

# SimScene-tier — `openral sim run --save-video` (debug smoke; no eval JSON).
just sim-libero                     # SmolVLA × LIBERO        (GPU + MUJOCO_GL)
just sim-xvla-libero                # xVLA × LIBERO           (Florence-2)
just sim-pi05-libero                # π0.5 × LIBERO           (≥8 GB VRAM)
just sim-act-libero                 # ACT × LIBERO            (paper protocol)

# RoboCasa has no dedicated recipe — drive it through the generic `sim-eval`
# with one of the in-tree RoboCasa SimScenes (XR-1 is the maintained pairing;
# there is no π0.5 RoboCasa rSkill).
just sim-eval scenes/sim/xr1_robocasa_pnp.yaml --rskill rskill://rskills/xr1-robocasa

# BenchmarkScene-tier — `openral benchmark scene --no-update-manifest \
#     --n-episodes 1 --save-dir` (paper protocol, single rollout for smoke).
just sim-metaworld --task metaworld/reach-v3
just sim-maniskill3                 # SAPIEN-backed PickCube-v1
just sim-simpler-widowx             # RLDX-1 × WidowX carrot-on-plate
just sim-act-aloha                  # ACT × gym-aloha bimanual cube-transfer
just sim-diffusion-pusht            # Diffusion Policy × gym-pusht (CPU)
just sim-custom                     # ACT × gym-aloha insertion (rskills/act-aloha-insertion)

just sim-audit runs tools/audit_sim_configs.py over the per-tier catalogue (one real rollout per scene × rSkill row) and classifies each row's outcome into a JSON report + Markdown table. just sim-eval <config> runs one scene config through openral sim run; suites run through openral benchmark run --suite <id>.

Sim Environments

Quick CLI

DeployScene catalogue (scenes/deploy/)

SimScene catalogue (scenes/sim/)

BenchmarkScene catalogue (scenes/benchmark/)

Justfile shortcuts

See also

DeployScene catalogue (`scenes/deploy/`)

SimScene catalogue (`scenes/sim/`)

BenchmarkScene catalogue (`scenes/benchmark/`)