Data for models that
move, play, and think.
Real-world training data for frontier AI models — not more text, but multimodal trajectories with physics, interaction, and embodiment.
The seven public samples below are representative slices Tidel has prepared for the robotics industry: real-robot teleoperated dual-arm and humanoid, primitive-level robotic arm manipulation labels, handheld-gripper (Hand UMI) demonstrations with vision-tactile fingers, tactile-glove hand capture, first-person view from a head-mounted stereo rig, and the outdoor delivery rider's perspective. Every sample is packaged in the same format we ship in production (synchronized multi-camera + depth + body pose + frame-level action labels).
tidelaccess
Real-robot teleoperation data
Operators control real robots in real time via exoskeletons, capturing the most complete, physically grounded sensor signals.
Wheeled dual-arm + UMI gripper
Wheeled dual-arm mobile robot · across 8 environment types — home / office / kitchen / warehouse, etc.
Humanoid + dexterous hands
Unitree G1 full-size biped · teleoperated in real time by an operator in an XR headset
Robotic arm manipulation
Dual-arm platform on a wheeled base · every sub-action attributed to the arm that performs it
Simulated / synthetic data
Generated in parallel in virtual environments like NVIDIA Isaac · infinite scale · Sim-to-Real gap.
Human motion capture (MOCAP)
Human trajectory data with motion capture, sensor gloves and handheld grippers, mapped precisely onto robots.
Hand UMI · handheld grippers + vision-tactile
A person holds two handheld grippers, one per hand, each with a camera-based tactile pad on both fingers · stereo egocentric headset · both grippers and the head tracked in one AprilGrid world frame
Tactile gloves + egocentric hand pose
Dual instrumented tactile gloves + head-mounted 4-camera rig · per-finger contact force at joint-level precision
Pure-vision egocentric data
First-person video without sensors / mocap, highly scalable, building general world representations.