R1 LITE / METHOD NOTE
From recordings
to testable assets.
Four real cameras, a manufacturer robot model, and an estimated scene reconstructed on one logical timeline.
The complete cup-nesting sequence follows recorded robot feedback and image-estimated cup trajectories. Native dynamics are tested separately below.
Three harnesses
Object: preserve the interaction
Three hollow cups use segmented collision geometry, positive mass and inertia. Their estimated height is 81.1 mm, mouth diameter 57.0 mm and mass 18 g. The hollow opening remains available for nesting.
Scene: align geometry and cameras
The manufacturer R1 Lite model, blue adapters, table and cups are assembled in metres. Camera views use nominal optics and image fitting. The upper room is a reference photograph projected onto an estimated shell; robot and cups are separate 3D objects.
Task: replay measured motion
Fresh arm and torso feedback drive the robot reconstruction. Cup trajectories use image observations and gripper poses. Source sensors are asynchronous; sharing a video timeline does not establish hardware exposure synchronization.
Six native checks passed
Isaac Sim / PhysX, 240 Hz, gravity 9.81 m/s². Thresholds and source hashes are in the numerical evidence summary ↗.
| Check | Tested condition | Result |
|---|---|---|
| Asset structure | Metre scale, positive mass, segmented hollow collision, central cavity clearance ≥ 5 mm | PASS |
| Effective inertia | Native mass and principal inertia remain positive | PASS |
| Free fall & settle | 10 cm release, 2 s simulation; final height error ≤ 4 mm | PASS |
| Bounded push | Force 1.2 × weight for 50 ms, then 2 s; displacement measured at 20.6 mm | PASS |
| Gravity nesting | Free upper cup / static lower fixture, 3 s; offset ≤ 10 mm, tilt ≤ 10° | PASS |
| Scene ready hold | One-second native initial-state hold; no pose or joint writes after attach | PASS |
The purple cup ran the dynamic tests. Red and green share its physical structure. The robot base is fixed and self-collisions are disabled.
What remains estimated
Cup dimensions, mass and friction; blue adapter geometry; table and room dimensions; camera intrinsics, distortion and exposure timing. The model is not calibrated against measured object materials or camera K/D.
The full robot task has not been executed as a dynamic contact rollout. The ready hold and cup tests do not establish autonomous grasping, nesting success or real-world policy transfer.
The four matched views use independent 10 fps renders and the studio overview 5 fps. Frames are repeated into 20 fps containers to preserve the 75.35 s source sequence.
Source & reuse
The three procedural cup files are available in the asset section. The robot geometry comes from Galaxea’s official USD repository ↗; robot geometry is not repackaged in this public page.
Raw telemetry, original archives and the complete project remain outside the public site. The native report is represented here by a numerical summary with source hashes.