FF
Bilateral force feedback
Contact forces at the follower arm are felt at the leader arm. The operator steers by feel, not just by sight.
Capability · Teleoperation
An operator drives. Yuri does the work.
Yuri is built to be teleoperated before it is autonomous. Force-feedback OpenLeader leader arms, Meta Quest 3/3S VR teleop, and a real-time operator-to-robot loop let a person do the job through the robot — and every run becomes a demonstration.
OpenLeader force feedback
A matched pair of leader arms, tuned for feel.
Every Yuri — Desktop or Mobile — ships with a matched pair of OpenLeader force-feedback leader arms mounted on their own stand. Teleoperate the follower arms and feel what the robot feels, because contact forces at the gripper are reflected back into the operator’s hands.
Force feedback is what makes Yuri different from a joystick-and-camera teleoperation rig. Precision tasks — inserting a vial, seating a screw, handling something fragile — are hard to do by sight alone. OpenLeader is designed so the operator can feel the contact and correct in real time, which is also what makes the resulting demonstrations high-quality training data rather than noisy remote-control footage.
2×
Matched bimanual pair
Two leader arms, one per hand, so both sides of a bimanual task move with natural coordination.
≡
Smooth, low-friction motion
OpenLeader arms are tuned for smooth, repeatable motion — precise intent, no fighting the hardware.
↺
Bilateral and unilateral modes
Run with force feedback on both sides, or detach one side for simpler tasks. Mode switching is part of the shipped stack.
Shipped model may differ from video.
Quest 3/3S teleoperation
Spatial control without a physical leader.
Yuri supports Meta Quest 3/3S teleoperation over the local network. The operator sees live wrist and head camera feeds inside the headset and drives the arms with their own hands — a fast way to capture demonstrations on tasks where a physical leader stand is not convenient.
Quest teleop is spatial and whole-body: instead of moving a physical leader arm, the operator moves their own hands in 3D space, and Yuri’s inverse-kinematics solver maps that to the follower arms. Round-trip latency on a local network is targeted under 10 ms, which is enough for contact-rich manipulation and stable force cues.
VR
Meta Quest 3/3S
Supported out of the box. Live camera feeds and arm pose streamed to the headset over local Wi-Fi.
IK
Inverse-kinematics mapping
The operator’s hand pose drives Yuri’s arm pose in real time. No programming — just motion.
<10ms
Local-network round trip
Sub-10 ms round-trip target on a local network keeps demonstrations stable and contact cues crisp.
3D
Spatial, whole-body capture
Record what the operator did in space, alongside joint state, gripper state, and camera streams.
Quest teleop of Yuri Desktop. Shipped model may differ from video.
Demonstration capture
The operator is the method. The demonstrations are the product.
Every teleoperated run is a demonstration, and the demonstrations are Yuri’s training data. A single teleoperation session records what the operator did, what the robot did, what the cameras saw, and what the gripper felt — all synchronized on a common timestamp.
That synchronization is the point. A video of the task is not enough to train a policy. What a Physical AI team needs is a dataset where the wrist camera frame at time t lines up with the joint state at time t, the gripper aperture at t, and the language label describing the task. Yuri records that automatically on every teleop run.
-
Operator intent
Leader arm pose (or Quest hand pose) — the "what should happen" signal, captured at full teleop rate.
-
Robot state
Joint angles, torques, gripper aperture, and (on Mobile) base state — the "what did happen" signal.
-
Camera streams
Wrist Arducam RGB frames, head RealSense depth frames, and optionally Quest headset views — the "what did it see" signal.
-
Language labels
Task descriptions attached to each episode so policies can be conditioned on natural-language commands.
The output is a LeRobot-format dataset, ready for imitation learning and ready to sync to the Hugging Face Hub. Full details live on the Robot Data Collection page.
Safety and fallbacks
Teleoperation that stays predictable.
Teleop is only useful if the operator trusts it. Yuri ships with a joystick/web fallback for when the leader arms or Quest are not available, a hardware e-stop that cuts motor power, and a depth camera on the head to detect people who are too close.
WEB
Joystick and web fallback
If the leader arms or Quest headset are unavailable, the included web app provides a joystick-based fallback for the arms and base.
E
Hardware e-stop
An emergency stop that cuts motor power — always one press away, on the robot and (on Mobile) on a wireless remote.
◎
Depth-camera safety
The head-mounted RealSense depth camera can detect people who are too close to the robot during a teleop session.
CAN
Low-latency control loop
CAN bus between host and arms, tuned for high-rate, deterministic control — force feedback stays smooth.
What teleoperation unlocks
Teleoperation is step one. Data and learning are the destination.
Teleoperation exists to produce demonstrations. Demonstrations exist to train policies. Policies exist to remove the operator. Each of those is its own capability on Yuri, and each links directly to the next.
Teleoperate Yuri in your lab.
Tell us what task you want to capture first and which configuration fits — Yuri Desktop or Yuri Mobile. We will follow up with pricing, availability, and a suggested demonstration plan.