5,000+
Motion clips
24 h
Cleaned motion
4
Professional actors, dancers, gymnasts, more.
120 FPS
Vicon optical capture
8
QA checks / clip
Transforming raw performance data into reliable physical ai deployment models should not require tedious manual cleanup. Powered by advanced video to motion capture and automated ai motion capture engines, our pipeline instantly bridges human motion to hardware skeletons using an intelligent auto rigger built for modern robotics labs.
DATASET OVERVIEW
Everything you need to evaluate, at a glance.
When you are training models across rt x, lerobot, and pi0 architectures inside simulators like isaac sim, telemetry depth matters. Whether you are benchmarking frames per second (fps) or evaluating embodied ai performance, our platform provides comprehensive data visibility.
DATASET VOLUME
5,000+
motion clips
24 h
of cleaned motion
4
performers
CLIPS BY CATEGORY · BREAKDOWN
Everyday & postural transitions 44%
Locomotion & agile navigation 24%
Yoga, floor & inversion work 16%
Expressive & dance 11%
Martial arts & sports 5%
CAPTURE
120 FPS
Optical marker-based capture of consented performers.
✓
Full body + both hands (fingers)
✓
Floor contact & transitions
✓
Continuous takes up to ~80 s
SKELETON & FORMATS
29 joints · Y-up · meters
LICENSE
Commercial
Training rights for production models. Research and custom-capture terms available.
Designed for leading robotics teams working at the intersection of artificial intelligence and robotics. Every trajectory is processed with clean skeletal parsing, auto rigging, and precise model calibration, ensuring collision-aware and deployment-ready workflows.
PIPELINE
From raw markers to training-ready files.
Raw mocap data is notoriously messy. Our pipeline transforms raw inputs into a clean training dataset using rigorous supervised learning pipelines, precise annotation, and automated key points extraction.
01Raw MoCap ingress
02Clean & filter
✓
Marker labeling
auto + manual
✓
Gap filling
occlusions
✓
Jitter filter
low-pass
✓
Foot-contact cleanup
no sliding
✓
Human QA
per clip
03Retarget
→
Auto bone mapping
source → target
→
Bind-pose normalize
T/A-pose
→
Collision avoidance
shoulder · elbow
→
Joint-limit clamp
per embodiment
04 OUTPUT FORMATS
BVH
motion hierarchy
FBX
DCC & engines
GLB
web / three.js
NumPy.npz
joint arrays for training
URDF joint angles*
robot-ready
* Robot / URDF outputs on request, to confirm with data team
ROI CALCULATOR
What clean motion data is worth to your team.
Plug in your own numbers. See how many engineering hours, dollars and sim-to-real failures you save by training on pre-cleaned, physically verified motion instead of building the pipeline in-house.
Your setup
LIVE ESTIMATEAdjust the inputs to match your team. Results update instantly.
Target robot configurations
Pre-calibrated for Unitree G1 (29 DoF). Custom URDF/MJCF models supported on enterprise request.
2 setups
1 setup
5 setups
Clips required per quarter
Quarterly motion dataset volume needed for policy training.
2,500 clips
500 clips
5,000 clips
Hours spent per week on manual data cleanup
Current team hours spent cleaning markers, tuning joint limits, and fixing foot skating.
60 h / week
0 h
120 h
Monthly hardware repair budget
Hardware costs from actuator burnout, joint over-torque, and collision damage.
$8,000 / mo
$0
$50k
Robotics / RL Engineer hourly rate
Fully loaded compensation rate (base salary + benefits/equity).
Your inputs stay in the browser — nothing is sent until you request a report.
YOUR ESTIMATED IMPACT
LIVE ESTIMATE
Engineering hours saved
323 h
/ month recovered
≈ 1.9 engineers back on policy work instead of mocap cleanup.
Training cost reduction
$37.5k
/ month
Engineering time + avoided hardware repairs.
Real failure risk
-68%
faults
Fewer boundary-collision and torque-limit faults.
Sim-to-Real Verification Standard
✓
Zero joint-velocity & torque-limit violations
✓
Ground reaction force & foot-contact aligned
✓
100% MuJoCo & Isaac Gym simulation-ready
Unverified in-house data
Krikey verified motion
Boundary-collision faults
Torque-limit faults
APPLICATIONS & CAPABILITIES
Powering the next generation of physical intelligence.
From reinforcement learning benchmarks to commercial humanoid manipulation, explore how pre-verified motion data accelerates robotics AI pipelines.
Embodied AI
Advanced models merge artificial intelligence and robotics to control dynamic hardware systems. These agents leverage physical intelligence to navigate unstructured data solutions in real-world environments.
Physical AI
Modern development pipelines integrate foundation models into physical ai frameworks to enhance spatial reasoning. This approach powers smart robot deployments across diverse operational settings.
Audio AI
Specialized audio ai architectures process complex sound streams to improve human-robot interaction. These systems complement computer vision models for multi-modal sensing.
AI Motion Capture
High-precision ai motion capture transforms optical tracking data into clean skeletal animations. This technology accelerates the creation of robust training datasets for robotic control.
AI Training Datasets
Curated ai training datasets provide the foundation for robust robotic perception and control. High-quality annotations ensure models generalize effectively from simulation to reality.
VLA Bounding Box
Vision-language-action models utilize vla bounding box annotations to ground spatial targets. Such precise data annotation is critical for successful manipulation tasks.
Multi Modal AI
This enables artificial intelligence robots, smart robot frameworks, and mobile robot deployments to scale efficiently across autonomous mobile robots, autonomous mobile robotics, and industrial mobile robots.