YAM Bimanual Teleoperation Data for Robot Learning
YAM arm data for robot learning, recorded by teleoperating YAM bimanual arms as ready-made kitchen, bedroom and staged retail sets or custom tasks. Each episode pairs camera views, joint states for both arms and commanded actions on one clock, delivered in LeRobot format with tactile or force-torque and depth as add-ons.
YAM Bimanual Arms and Their Leader Arms
YAM is a 6-DOF tabletop arm made by I2RT, driven over CAN bus and paired with a matching leader arm for teleoperation. It is the hardware behind our YAM arm data. The specs below are the maker's own, as listed on its store and documentation pages in November 2026.
Arm and gripper
The maker lists the YAM at a 2 kg nominal payload, a 750 mm maximum workspace range and a 95 mm gripper throw. The Standard, Pro and Ultra tiers share one SDK, and BIG YAM is listed for heavier payloads.
Leader arms and the bimanual cell
The leader arm is kinematically identical to the follower, with a teaching handle in place of the gripper. The follower mirrors the handle in real time, and a trigger opens and closes its gripper. The maker's YAM Cell pairs two leader arms with two followers.
Software and control
An MIT-licensed Python SDK runs the arms, with MuJoCo models, gravity compensation and a record-and-replay example. Cameras are sourced separately, so camera placement is set by whoever builds the rig.
We record on YAM bimanual arms, with camera views and streams agreed per program. Our OpenArm dataset options cover OpenArm-class arms on the same ready-made and custom paths.
Ready-Made YAM Sets for Kitchen, Bedroom and Staged Retail Work
Ready-made sets are YAM episodes we have already recorded, grouped into three task families and shipped with per-episode provenance and consent records.
Kitchen set
Bimanual YAM episodes recorded in kitchen settings.
Bedroom set
Bimanual YAM episodes recorded in bedroom settings.
Staged retail set
Real teleoperation recorded on YAM arms in a staged mock store set.
Non-exclusive license
Other teams can license the same episodes, and you can sublicense them to your own client.
LeRobot delivery
Delivered in LeRobot format, with MCAP, HDF5 and RLDS on request.
Sample pack on request
A representative sample pack with the same modalities, calibration and delivery format you'd get in production.
Custom YAM Collection for Your Tasks
When your evaluation needs tasks outside the ready-made families, we record them to your task list, written as the objects, actions and success conditions your policy is evaluated on. Custom collection runs on one of two paths, with add-ons on either:
A new task on YAM, or a new robot.
Both paths run on the same pipeline, from a pilot you evaluate first to human verification against agreed acceptance criteria and LeRobot delivery. The path you choose sets which robot records the data and how the pilot is scoped.
Ready-Made YAM Sets and Custom YAM Collection Compared
Both deliver YAM episodes in LeRobot format and differ in who sets the tasks, which streams you can add and how the data is licensed.
| Option | Ready-made YAM sets | Custom YAM collection |
|---|---|---|
| What you get | Episodes already recorded on YAM bimanual arms | New episodes recorded to your task list |
| Tasks | Kitchen, bedroom and staged retail families | Your tasks, objects and settings, on YAM or a new robot or embodiment |
| Streams and add-ons | The streams recorded in each set | Core streams agreed per program, plus tactile or force-torque and depth on request |
| How you evaluate it | A representative sample pack on request | A pilot you evaluate before scaling |
| Format | LeRobot, with MCAP, HDF5 and RLDS on request | The same |
| License | Non-exclusive, and you can sublicense it to your own client | Non-exclusive or exclusive, with exclusive terms priced above non-exclusive |
What Each YAM Episode Records
Each episode is one synchronized recording of a two-handed task, from what the cameras saw to what the operator commanded, with every stream on one clock. Streams are agreed per program.
Camera views
Wrist and scene views, the close-up at each gripper and the wider workspace.
Joint states for both arms
Joint positions for the left and right arm at every step.
Gripper states
Each gripper's state alongside the arm joints, so grasp timing is in the record.
Commanded actions
The commands sent to both arms at each step, the targets a policy learns to reproduce.
Teams training ACT or Diffusion Policy can compare these streams with what each model expects on our page on ACT and Diffusion Policy training data. Our teleoperation and capture services page covers the other robots we record on.
Add-On Streams for YAM Episodes
Two add-ons extend the core streams when your task depends on contact or distance.
Tactile or force-torque
Tactile or force-torque streams record contact and force during each grasp, on the same clock as the cameras and joints. Our tactile manipulation data page covers that capture in detail.
Depth
Metric per-frame depth, so positions and distances in each episode come out at true scale.
Add-ons are agreed in scoping, and the pilot carries the same streams you get at scale.
Delivery Formats and Licensing
YAM episodes load into the training stack your team already runs, with licensing set per option.
LeRobot by default
Camera streams, joint states and actions arrive in LeRobot format, ready for policy training.
MCAP, HDF5 and RLDS on request
For loaders, replay tools or storage that expect them, agreed with you in scoping.
Ready-made license
Ready-made data is licensed non-exclusive and can be sublicensed to your own client, which suits teams that build models for others.
Custom license
Custom collection runs on non-exclusive or exclusive terms. Exclusivity covers the data and its resale rights, separate from the robot that recorded it. Exclusive terms are priced above non-exclusive, and term and scope are agreed once a pilot is defined.
Every episode carries provenance and consent records, with signed participant and site agreements permitting commercial training and a chain of custody you can audit.
What Shapes the Engagement
We scope every YAM program against these factors with you, then quote the program as a whole.
Ready-made or custom
A ready-made set, or custom collection on one of the two paths.
Robot and gripper
YAM bimanual arms, or a new robot or embodiment for your policy.
Tasks and settings
The task list, objects and settings, written as your policy's success conditions.
Streams and add-ons
Camera, joint and action streams, plus tactile or force-torque and depth where needed.
Acceptance and QC criteria
What counts as a reject, such as a dropped object, a collision, a stream out of sync or an incomplete task.
Pilot and license terms
What the pilot covers for your team to train and evaluate on, and the license terms that follow.
Have a YAM Task to Record?
Send us the task and the streams your policy needs. We'll scope the work and send a representative sample pack with the same modalities, calibration and delivery format you'd get in production, so your team can inspect sensor alignment before any commitment.
Book a Strategy Session →Frequently Asked Questions on YAM Arm Data
YAM arm data already recorded on bimanual YAM arms, grouped into kitchen, bedroom and staged retail task families. Sets ship in LeRobot format, with MCAP, HDF5 and RLDS on request, and are licensed non-exclusive, so you can sublicense them to your own client. A representative sample pack is available on request.
The staged retail set is real teleoperation of YAM arms in a mock store set built for recording, so every camera frame, joint state and action in it comes from the real arms.
We scope and quote each program as a whole, against the factors above. Ready-made sets are licensed non-exclusive. Custom collection runs on non-exclusive or exclusive terms. Exclusivity covers the data and its resale rights, and exclusive terms are priced above non-exclusive. Term and scope are agreed once a pilot is defined.
Related Physical AI Solutions
Teleoperation and Capture Services for Robot Learning
Learn more →ACT and Diffusion Policy Training Data for Robots
Learn more →OpenArm Dataset Options: Ready-Made and Custom
Learn more →UMI Data Collection for Robot Learning Teams
Learn more →Embodied AI Training Data: Demonstrations for Robot Policies
Learn more →Tactile Manipulation Data: Tactile Sensing and Force Arrays for Robots
Learn more →Humanoid Robot Training Data and Teleoperation
Learn more →Physical AI Training Data: Multimodal Field Capture for Robot Learning
Learn more →Physical AI and Robotics Consulting, Development and Training Data
Learn more →Physical AI and Robotics Consulting
Strategy, data capture and model fine-tuning help for bimanual arm programs, from one team.
See physical AI services →AI Strategy Session
A 30-minute scoping call on your YAM tasks, streams, add-ons and delivery format.
Book a session →Tactile Manipulation Data
Tactile and force streams for contact-rich manipulation, synced to vision.
See tactile manipulation data →Scope Your YAM Arm Data
A 30-minute strategy call. Tell us the tasks, settings and streams. We'll scope a ready-made set or a custom YAM pilot with the format and acceptance criteria your team needs.