LIVUMI GRIP
Capture grasping and manipulation in a robot-like end-effector form
Fuse vision, pose, gripper and tactile data so high-quality robot demonstrations begin with real interaction.
ONE-LOOP PIPELINE
One-loop collect-train-deploy pipeline
From human-centered capture and data governance to VLA / VAM training, deployment and feedback, LivSyn builds the devices, edge platform, pipeline and models as one system.
01
Capture
UMI · EGO · DexGripper, headset and glove form factors collect high-quality demonstrations without requiring a robot body.
02Govern
DataCube · LDPFrame-level synchronization and inspection with traceable cleaning, deduplication, compliance and versioning.
03Annotate
LDPMultidimensional annotation and DataViewer previews produce traceable, training-ready datasets.
04VLA / VAM Training
RUDA · PhiAgentA unified architecture trains vision-language-action and vision-action models across embodiments.
05Deploy
RoboAgentCompile policies and send motion plans to robots for stable execution and closed-loop feedback.
06Data Feedback
LeRobot v2.1 · RLDSReal-world execution flows back into the pipeline for continuous model iteration and standard export.
DATA SAMPLES
See the data quality first
These are raw replays from real collection scenarios. Visit Data Samples for full datasets, schemas and downloads.
[Ego] Parcel handling · First-person view
[UMI] Picking fruit · Gripper view
PARTNERS & CLIENTS
Partners / Clients
