Open Navigation

← Explore the warehouse

Open Navigation warehouse replay

One complete original warehouse and its source forklift, with a recorded first Nav2 mission. Both navigation legs and the source handling waits completed. The workload benchmark's throughput measurements, optional vision-language model, and other missions are outside this recording.

Original scene and mission

The pinned Open Navigation source supplies the warehouse, robot and mission dispatcher. The source lock identifies 441 Git blobs, the required Fuel cache and aliases. The full native scene contains 1,388 models, 1,405 links and 2,249 visuals. Its 28 unique mesh, material and texture files total 40,168,724 bytes. Original vendor geometry loads from pinned upstream URLs.

ROS Jazzy and Gazebo Sim 8.15.0 run the original navigation stack. The observed physics engine is DART with 10 ms steps; the source profile's “tpe” name is not treated as an engine identity. Dispatcher seed 42 selects pickup (−8, −52.55) m and drop-off (−35.5, −36.7) m. Both outer navigation legs succeed on their first attempt; the original navigation stack performs internal recovery. Each handling phase is a ten-second wall-clock wait, not physical pallet manipulation.

The native mission receipt and source and geometry audit preserve that outcome. There is no source episode-step horizon to invent from the sample count. The recorded “terminated” field marks completion of this mission recording, not a Gym termination signal. The optional model server was absent; no model inference or simulated hardware-throughput result is claimed.

Measured replay and camera

The episode manifest has 11,847 timestamped samples of 19 body frames over 191.81 seconds. Joint positions, joint velocities and per-step applied actions were not captured as aligned channels. Original Twist messages retain their own header and reception timestamps as observations. Body positions and rotations are interpolated only between the surrounding recorded samples.

The front-camera clip preserves all 960 observed images at 640 × 480. Its frame index retains each native timestamp, raw RGB hash and encoded presentation timestamp. H.264/YUV420 is a lossy derivative, not pixel-identical RGB. Every encoded frame and timestamp was checked; minimum measured quality is 46.03 dB PSNR against the original images converted to YUV420. The original PNG paths in the index identify local audit inputs, not additional served files.

The first image appears 0.01 seconds after the first body sample. Playback holds the latest observed image with its source time and age, including the final short interval after the last image at 191.78 seconds. A browser frame callback must confirm the requested image before the paired body frame is displayed. The movable 3D inspection camera is separate from the native sensor view.

Desktop scope and loading

The desktop component check covers 41 seeks, body/camera timing, playback, looping, obsolete requests, graphics-context recovery and resource disposal. All 19 displayed body origins match their source coordinates within 2 micrometres at the checked selections. Broader native geometry verification compares original mesh attributes and 900,372 body/basis points. These checks do not establish identical native rendering or live-network performance.

Original geometry is loaded by spatial region and repeated meshes are instanced. The scene has 63 finite regions; visited regions share a bounded 28-file source closure until the world is closed. No mesh simplification or source-provided LOD is claimed. The hardware WebGL camera check uses software video decoding. Default headless Chrome also passed functional camera checks with software 3D rendering. Forced hardware-decoder overrides failed in the test workspace; hardware decoding is not an acceptance claim. Physical-phone testing is deferred.

Source attribution and files

Open Navigation source code is Apache-2.0. Asset terms are separate, including noncommercial and no-derivatives restrictions. Public accessibility is not a blanket licence grant. The source distribution's model versions and notices are retained below and in the scene descriptor; original geometry is neither converted nor rehosted by this viewer.

Original model attribution from the pinned source and Fuel model metadata.
ModelContributorSourceDeclared licence
Tugbot-charging-stationMovAiVersion 2Creative Commons Attribution Non Commercial No Derivatives 4.0 International
Casual femaleOpenRoboticsVersion 4Creative Commons Zero v1.0 Universal
ForkliftOpenRoboticsVersion 1Creative Commons Attribution 4.0 International
palletMovAiVersion 2Creative Commons Attribution Non Commercial No Derivatives 4.0 International
pallet_box_mobileMovAiVersion 3Creative Commons Attribution Non Commercial No Derivatives 4.0 International
shelfMovAiVersion 1Creative Commons Attribution Non Commercial 4.0 International
shelf_bigMovAiVersion 2Creative Commons Attribution Non Commercial No Derivatives 4.0 International

Full scene and episode registry · scene assembly and original asset URLs · observed commands · native runtime and reproduction notes.