sentientx.

SERVICES

Four Pillars. One Partner.

We collect the real-world training data your robot needs to deploy: teleoperation, motion capture, UMI gripper data, and real-world RL. No simulation. No egocentric video. Just high-value data that transfers.

Teleoperation Data Collection

Direct robot control by human operators, capturing full action sequences with synchronized sensor data.

What You Get

  • End-effector poses (x, y, z, roll, pitch, yaw) + gripper state
  • 10-50 Hz recording frequency (configurable)
  • Full proprioceptive state (joint positions, velocities, torques)
  • Multi-view RGB (wrist, third-person cameras)
  • Depth and point cloud data (optional)
  • Language annotations (task descriptions, subtask segmentation, hindsight labels)

Output Formats

RLDS, HDF5, LeRobot, Open X-Embodiment format, or custom schema

Differentiator: Domain-expert operators recruited for your robot's target application. Not generic teleop workers.

Typical Engagement

Pilot batch of 100-500 demonstrations, then scale to 5,000+ based on model performance.

Motion Capture Data Collection

High-precision tracking of human movement for humanoid locomotion and dexterous manipulation.

What You Get

  • Full-body skeletal tracking
  • Hand and finger pose capture
  • Sub-millimeter precision (optical systems)
  • Synchronized video streams
  • Retargeting to your robot's kinematic model

Use Cases

  • Humanoid locomotion training
  • Dexterous hand manipulation
  • Human-robot interaction modeling

Facility

Dedicated motion capture studio with optical and inertial systems.

UMI Gripper Data Collection

Universal Manipulation Interface data for precise dexterous manipulation tasks.

What You Get

  • High-fidelity gripper pose and force data
  • Multi-finger coordination trajectories
  • Contact and grasp force measurements
  • Object interaction sequences
  • Synchronized visual and proprioceptive streams

Use Cases

  • Dexterous manipulation
  • In-hand object reorientation
  • Precision assembly tasks

Differentiator: UMI-compatible data collection setup with calibrated force sensing and multi-view cameras.

Real-World Reinforcement Learning

Robot self-play with reward signals, enabling autonomous improvement from deployment experience.

What You Get

  • Supervised autonomy: We run your robot in our facility with human supervision
  • Reward labeling: Human annotators provide reward signals for autonomous trajectories
  • Recovery demonstrations: When the robot fails, we capture human recovery trajectories
  • Edge case collection: Deliberate stress-testing of policy boundaries

Key Insight: The final 10% of performance comes from learning to recover from failures. We capture the data that teaches your model robustness.

Complex Task Decomposition

For industrial applications (shipbuilding, welding, heavy manufacturing), we design modular training approaches that break complex workflows into trainable components.

Example: Shipyard Welding Robot

Instead of requiring an actual shipyard, we design a modular training rig that simulates:

  • • Confined space navigation
  • • Multi-joint welding sequences
  • • Variable surface angles
  • • Environmental conditions (heat, fumes simulation)

Each component is trained separately with perfect repeatability, then composed for full-task deployment.

You don't need to ship us to a shipyard. We bring the shipyard-equivalent to our facility.

Hardware & Integration

Option 1: You Ship Us the Robot

  • • We set up in our facility
  • • 1-2 weeks for integration and operator training
  • • You retain remote access for monitoring

Option 2: We Deploy to You

  • • Our team on-site at your facility
  • • We manage operators and collection
  • • You retain full control

Supported Hardware

Platforms

  • • Humanoid platforms (most major vendors)
  • • Mobile manipulators
  • • Industrial arms
  • • Dexterous hands
  • • Custom hardware (scoped separately)

Teleoperation Systems

  • • MyArm MC kits (2-week delivery)
  • • UMI gripper setups
  • • Custom leader-follower setups
  • • VR teleoperation (Quest, Vision Pro)
  • • Keyboard/spacemouse interfaces

Data Formats

  • • RLDS (TensorFlow Datasets)
  • • HDF5
  • • LeRobot format
  • • Open X-Embodiment
  • • Custom formats on request

Work With Us

Tell us about your robot and your data needs.