Teleoperated bimanual episodes on the YAM platform, across two capture rigs. Every episode ships synchronised joint telemetry in MCAP with the video. The clips below show the overhead camera only. Each episode also carries both wrist views.
Tasks whose episodes are uniformly brief — a single grasp-and-place loop, repeated. Useful when you want many independent demonstrations per hour of data.
Pick a hair dryer off the table and hang it on the wall hook.
ZED rigStereoMCAP
Transfer loose fruit into a basket, one piece at a time.
RealSense rig+ DepthMCAP
Lab-bench handling: well plates, petri dishes, sample trays and lids moved between stations.
ZED rigStereoMCAP
Read the contents of each bottle and sort it into the matching bin.
ZED rigStereoMCAP
One to two minutes of continuous bimanual work per episode. The bulk of the corpus sits here.
Clear dirty dishes from the table into the sink, then push the chairs back in.
RealSense rig+ DepthMCAP
Precision placement of small parts into a pegboard grid — the tightest tolerances in the corpus.
ZED rigStereoMCAP
Unpack a bulk bag of parts and repack them into their original moulded box.
RealSense rig+ DepthMCAP
Bimanual deformable-object manipulation: thread, cross and pull a lace through to a finished knot.
RealSense rig+ DepthMCAP
Multi-stage episodes running several minutes, where the arms have to sequence sub-tasks and recover from their own earlier choices.
Clear a cluttered desk: bin the rubbish, return everything else to where it belongs.
RealSense rig+ DepthMCAP
Measure parts with a digital height gauge and sort them by how close they land to a target dimension.
RealSense rigMCAP
Work through a pile of T-shirts, sweaters and socks, folding each one away into a drawer.
RealSense rigMCAP
Full-length folding runs: shake out, flatten, fold and stack.
ZED rigStereoMCAP
Both rigs run the same YAM dual-arm platform with three cameras and produce the same MCAP telemetry; they differ in the camera hardware and so in what the video gives you.
| ZED rig | RealSense rig | |
|---|---|---|
| Cameras | 3 × Stereolabs ZED | 3 × Intel RealSense D405/D435 |
| Overhead / environment view | cam_high | cam_high |
| Wrist views | cam_left_wrist, cam_right_wrist | cam_left, cam_right |
| Stereo video | Yes — both eyes kept, on all three cameras | No — single RGB view per camera |
| Depth | Recoverable from the stereo pair; not shipped as its own stream | Aligned depth video alongside the RGB |
| Synchronised telemetry | MCAP | MCAP |
| Raw capture format | Stereolabs .svo2 | h264 .mp4 per camera |
| Robot description (URDF) | yam.urdf | dual_yam_platform_d405_d435.urdf |
Previews are reduced-resolution. Full-resolution video, the second eye where the rig records one, and the MCAP telemetry are delivered separately.