Beyond motion.
The feel of the task.
Scenes and movement. Force, torque, and touch.
Commands, stiffness, and gaze that reveal intent.
One demonstration, shaped into experience a robot can learn.
One moment. More ways to understand it.
Conceptual visualization, not measured waveforms. Available channels and acquisition methods are defined for each project.
What should your
robot learn next?
Start with a task. Define the senses and variations that matter. This catalog helps shape the scope of acquisition.
Read the misalignment. Capture the fit.
Distinguish search, first contact, rising force, full seating, and retreat. Vary part tolerances and starting poses to record both clean fits and catches.
Related field tasks
- Pipe-spool flange alignment and gasket seating
- Hose insertion, clamping, and routing
- Cable laying, routing, and connector mating
- Wiring harness routing, clips, and connectors
This catalog describes acquisition scope, not ready-made dataset inventory, hours, or episode counts. We’ll confirm available data and additional acquisition plans with you.
What happened. What was intended.
From what moved to what was felt and how the operator intended to respond. Dream Set’s additional contact and intent modalities are highlighted in orange.
The scene
Video of the objects, tools, and workspace. The context around every contact.
Both wrist views
Close-up views from each wrist, connecting the objects in both hands with contact areas that might otherwise be hidden.
Robot motion
Joint states, end-effector position and orientation, and gripper motion. A record of what the robot actually did.
Applied force
The forces behind pushing, pulling, and supporting. Direction and magnitude, connected to movement and contact stages.
Applied torque
Moments during rotation, fastening, and twisting. Recorded separately from force, with coordinate frames and measurement locations.
Fingertip contact
How contact is distributed and changes at the fingertips. Robot-hand sensing and operator-glove sensing remain distinct.
Operator commands
The intended movement: where to go and how. Operator commands are recorded separately from the robot’s actual response.
Intended force and stiffness
How much force to apply. When to stay firm, and when to yield. The operator’s adjustments, connected to motion and contact.
Operator attention
Eye tracking connects attention to objects and contact points. Gaze is recorded as attention information, distinct from head direction or camera view.
The details that make data usable.
| Data element | Configuration principle |
|---|---|
| Timing | Preserve original sensor rates and timestamps, and document synchronization and resampling methods. |
| Physical meaning | Connect coordinate frames, units, mounting locations, and calibration history. Distinguish measurements from estimates and annotations. |
| Task context | Define objects, materials, tolerances, initial states, task stages, and operator attention. |
| Success and recovery | Label success, catches, retreats, regrasping, and stops separately, with evaluation criteria. |
| Quality and splits | Flag missing, saturated, or anomalous signals. Design training and evaluation splits by operator, part, and condition. |
| Delivery scope | Agree on formats, permitted use, quantities, and additional acquisition requirements for each project. |
These are design principles for data configuration and delivery. Each project’s actual specifications are documented in its data card.
Tell us what your robot needs to feel.
Share the task and your current robot and sensor setup. We’ll design the path from Dream Catcher acquisition to the data your project needs.