magicbio-demo

Demo videos and key images for the industrial autolab simulator (MagicSim: Device System + Matter System).

• routine/demoNN/ - routine track (one fixed arm + one device per hub)
• error_recovery/err_demoNN/ - error-recovery track (accident, then Vega + Sharpa restores the lab)
• teaser/ - teaser hero image (teaser_hero_3200x1800.png)

Each folder: *.mp4, kf_*.jpg key frames, sheet*.jpg contact sheet, captions.txt where available.

Routine videos

Five existing Isaac Sim demonstrations. Processing waits are shortened or time-compressed as described below.

Demo	Video	Duration	Key images
demo01	Conveyor, buffer station and plate sealer	50.8 s	Contact sheet
demo02	Balanced centrifuge loading and unloading	71.7 s	Contact sheet
demo03	Opentrons Flex liquid handling and PCR	90.7 s	Contact sheet
demo04	Stacked two-chamber incubator (v2)	167.7 s	Contact sheet
demo05	Six-vial freeze-drying batch	100.5 s	Contact sheet

Demo 4 links to the 1080p v2 video and matching v2 key images. The delivery manifest records video hashes, complete decode checks, keyframe timing checks and known limitations. These are demonstration clips, not repeated reliability measurements.

Folders currently uploaded (the set grows as demos finish):

• error_recovery/err_demo01
• error_recovery/err_demo02
• error_recovery/err_demo03
• routine/demo01
• routine/demo04: One continuous four-plate incubation run inside the aluminium/glass hub. The same arm opens and closes both the outer and inner cabinet doors using their handles. Unloading places the plate on a staging support before the empty hand closes the doors, then retrieves the plate for delivery. Manual hinges use an ideal constraint calculated from measured closed-gripper poses; friction-only contact is not validated. Cabinet panel contacts are suppressed during manual approach, manipulation and exit; collision-free planning is not validated. Native plate grasp and placement helpers are retained. Playback is 8x, with labelled 64x idle intervals; culture and incubation clocks use the documented demo compression. Carried plates use a pose lock acquired only after a measured closed-jaw grasp and verified physical lift; plate contacts are suppressed until the jaws reopen. Friction-only carrying is not validated. A four-second final still reuses the last recorded scene and shows the verified four-plate completion result. All 40 manual operations were visually reviewed in the four door-contact sheets (contacts_01.jpg through contacts_04.jpg). The filmed run completed, but its optional native trajectory export exceeded the recorder's 600-second close wait and remained uncommitted. The separate full four-plate validation probe committed one trajectory.
• routine/demo05: One new continuous full laboratory simulator run inside the aluminium/glass hub. Demo02's native UR10e and Robotiq 2F-85 use the original black dryer cover_handle, with no added yellow bail. Closed jaw gaps/TCPs are calibrated for the 19 mm original handle and 28 mm vials; acquisition and carrying are checked against measured pad surfaces. The robot performs four handle operations, loads and freeze-dries six vials, retrieves all six into the destination rack, and closes the lid. The verified final scene is held for two seconds. Playback is uniformly 4x. The process model runs at 2400 process seconds per simulated second, equivalent to 9600 process seconds per video second during the cycle. The entire compressed cycle remains visible without additional frame dropping. Lid manipulation uses an ideal constraint calculated from measured closed-gripper pad poses; friction-only manipulation is not validated. Vial pose-lock carrying and support docking stabilization are retained; robot-to-lid/handle and vial contacts remain enabled. A collision-free motion planner is not validated; native geometry, grasp clearances and dynamic contacts are checked. A private sample vial (49.5 mm high, 28 mm outer diameter) holds the unchanged 3 mL sample of 5% sucrose. Its collider clears the original lid by at least 4.5 mm throughout the 0–110 degree sweep.
• routine/demo03: One continuous complete 44-step Flex liquid-handling and PCR workflow with eight programmed PCR cycles inside a compact 1.34 x 1.57 m aluminium/glass hub. The brighter warm-white, grey and blue laboratory contains organised static equipment and supported supplies. Demo3-local translucent plate optics and CAD-conformal liquid display meshes separate refracting/contact surfaces while preserving the original Matter amounts, liquid/thermal models and physics. Calibrated robot, equipment and labware positions are retained; handling, deck and PCR views include a persistent full-enclosure overview. Every frame comes from the same continuous run at uniform 4x playback. The existing automation retrofit, device and Matter models are retained; thermal timing uses the shortened demo model and colour changes remain difficult to see behind the closed thermocycler lid.
• routine/demo04 (4 plates, different temperature/time, stacked two-chamber CO2 incubator) - v2, 1920x1080, full-frame Isaac Sim RTX real-time render (sampled lighting, GI, AO, DL denoiser, ACES tone mapping; the earlier grainy v1 was replaced). The whole picture and the top-right close-up camera are real render frames. Overlays drawn afterwards: the bottom status strip (time, current skill, per-plate state / time left / turbidity %), the inset border, and the "TIME-LAPSE x2.5" (x5 while idle) and "incubation times shortened (10-70 s)" labels. Incubation times are symbolic (A 37 C/70 s, B 30 C/12 s, C 10 s, D 16 s; plate A deliberately longer so the due order B, C, A, D differs from the arrival order). Yeast growth is an accelerated proxy (matter_time_scale 120), so turbidity differences are subtle on screen and mainly read from the strip. Outlets are 4 labelled carriers on one belt (out_30 / out_37 are logical labels). The incubator asset is a patched copy of a scanned asset; the background is the generic dreamLab lab. Disturbance hooks (external force, passing person) were only tested in non-simulator unit tests. The v1 render is in the repo history.
• routine/demo02: One continuous rack-to-centrifuge-to-rack run inside the aluminium/glass hub. The conveyor brings blue suspension in and reverses to send the rack outside after processing. UR10e with native Robotiq 2F-85 grasps the handle to open the lid before loading and after the rotor stops; the existing arm hook closes and latches it. Workbench references are fixed static colliders; inherited dynamic rigid bodies are disabled on the bench and visual-only lab neighbours. Before/after closeups show uniform blue suspension becoming clear upper supernatant and blue lower sediment through the existing Matter model. Video playback is uniformly 2x, with 200x Matter time only during closed-lid centrifugation; visible rotor speed is 1/50 of device RPM. Only the first .32 seconds of material warmup are trimmed, before the belt starts; the full ingress and operation remain. Opening uses an ideal constraint from measured closed-gripper poses; friction-only opening and collision-free planning are not validated. Original one-sided pad guidance for closure and tube extraction/return grasp weld are retained. The generic coloured material and cylindrical liquid section are visual approximations, not a calibrated assay. Laboratory visual refresh: soft white walls, cool grey floor, teal preparation cabinets and supported background equipment; suspended scan clutter removed and workstation props corrected to the desk height. The calibrated workcell, robot, device/Matter settings and physics clocks are retained. Native MoveL logged 0 timeout(s) and 2 stall(s); subsequent measured workflow checks passed, and the continuous footage retains those pauses.
• routine/demo01: UR5e with plate gripper; conveyor carrier -> buffer hotel -> sealer -> carrier. The upright station is a plate hotel with horizontally supported plates, rather than a plate standing on its edge. Buffer dwell is 10 s and sealing dwell is 4 s. Camera views are Isaac Sim output; labels, timers and view composition are post-processing. This is a single demonstration, without recorded disturbance or repeated reliability trials.
• routine/demo05: UR5e with Robotiq; six vials loaded, freeze-dried and unloaded in a laboratory cell. Transport uses pose locks and docking onto support surfaces rather than a verified friction grasp. Device processing is compressed by 2400x (1 s of simulated processing represents 40 min of process time), labelled in the film. Temperature, sublimation and mass-transfer parameters are engineering assumptions, not experimentally calibrated. Pressure is supplied by a demo-local chamber bridge. The source rack is obscured by the arm in the fixed cameras. The HUD and camera inset are post-processing. The reported successful batch is not a repeated reliability result.

Routine demo02 was re-recorded inside an aluminium/glass RAC-style hub. The canonical video and all ten key images now show the enclosed workstation (53.8 s, one uninterrupted complete run). The previously documented lid-contact and tube-extraction model assists remain and are labelled in the video.

Current demo03 revision — Stable translucent PCR plate and refreshed professional laboratory. One continuous complete Flex PCR run with eight programmed PCR cycles in the compact 1.34 x 1.57 m hub and a refreshed warm-white, grey and blue laboratory. Demo3-local frosted translucent plate optics, clear well walls and volume-preserving CAD-conformal liquid display geometry avoid visual surface overlaps. The Matter volumes, chemical/thermal model, physics and calibrated workstation positions are retained. Inspection camera framing is reviewed against actual equipment motion. Uniform 4x playback.

Current demo02 revision — Refreshed laboratory with fixed workbenches, conveyor and UR handle opening. One continuous run: the suspension rack enters the hub, the UR10e with native Robotiq grasps and opens the closed lid before loading, closes it for a balanced spin, grasps and opens it again after the rotor stops, returns the separated tube and conveys the rack outside. Fixed bench and background references disable inherited dynamic rigid bodies. The laboratory backdrop now has soft white walls, cool grey flooring, teal cabinetry, supported preparation equipment and softer ceiling lighting. Unsupported scan clutter was removed and the workstation monitor/keyboard are seated on the actual desk top.

Current demo05 revision — Native UR10e + Robotiq, original dryer handle and physically fitting short vials. One new continuous full laboratory simulator run inside the aluminium/glass hub. Demo02's native UR10e and Robotiq 2F-85 use the original black dryer cover_handle, with no added yellow bail. Closed jaw gaps/TCPs are calibrated for the 19 mm original handle and 28 mm vials; acquisition and carrying are checked against measured pad surfaces. The robot performs four handle operations, loads and freeze-dries six vials, retrieves all six into the destination rack, and closes the lid. The verified final scene is held for two seconds. Playback is uniformly 4x. The process model runs at 2400 process seconds per simulated second, equivalent to 9600 process seconds per video second during the cycle. The entire compressed cycle remains visible without additional frame dropping. A private sample vial (49.5 mm high, 28 mm outer diameter) holds the unchanged 3 mL sample of 5% sucrose. Its collider clears the original lid by at least 4.5 mm throughout the 0–110 degree sweep.

Current demo04 revision — Robot-operated incubator doors and staged plate transfers inside hub. One continuous four-plate incubation run inside the aluminium/glass hub. The same arm opens and closes both the outer and inner cabinet doors using their handles. Unloading places the plate on a staging support before the empty hand closes the doors, then retrieves the plate for delivery. A four-second final still reuses the last recorded scene and shows the verified four-plate completion result.