Rerun Recording

Rerun is a multimodal visualizer for robotics data. It logs 3D point clouds, meshes, camera frustums and images, transforms (/tf), and per-signal time-series onto a single scrubbable timeline — and it can also render 3D Gaussian-splat reconstructions (the splat point cloud / scene geometry). That makes Rerun a natural viewer for both a live robot run and the Scene Reconstruction capture of the workspace: you can scrub the manipulation episode and inspect the reconstructed environment with the same tool, in the same 3D view.

This page embeds an interactive recording of a full AR4 pick-and-place run in the tabletop Gazebo world — the 3D arm (URDF meshes driven by /tf), the wrist-camera stream, per-joint position plots, and the traced end-effector path, all on one scrubbable timeline.

How it is produced

The recording is captured by a decoupled recorder (ar4_rerun) that runs in its own container and talks to the ROS side only over Zenoh (zenoh-bridge-ros2dds). This keeps rerun-sdk (numpy 2) fully isolated from the ROS 2 Jazzy images (numpy 1.x). The resulting .rrd is uploaded to the artifacts bucket and embedded below via the pinned app.rerun.io web viewer (version matched to the SDK that wrote the file).

See ar4_rerun/README.md for the full record sequence.

The viewer loads the recording from https://ar4-physical-ai.aegeanai.com/rerun/ar4-pick-place/ar4_pick_place.rrd. The viewer version (0.29.2) is pinned to the rerun-sdk that wrote the .rrd — a newer recording will not load in an older viewer.