Every stage of robot skill training, in one platform.
Data collection, visualization & curation, training, and deployment share one strongly-typed data model and one version history, so a policy in production always traces back to the exact demonstrations that made it.
Stream any robot signal with one line of Python.
A log_* call for every one of 14 first-class modalities, joints, poses, grippers, RGB, depth, point clouds, language, and custom signals, each timestamped and synchronized the moment it lands.
- Live WebRTC monitoring and durable recording from every single call
- Resilient Rust/Python daemon: resumable chunked upload, offline mode, dual H.264 encoding
- URDF / MJCF robot setup with multi-instance and distributed teleop collection
Turn raw logs into training-ready datasets.
Merge heterogeneous traces onto one timeline, explore episode diversity visually, and curate with real tools, not scripts. Every training run points back to the exact synchronized snapshot it used.
- Multi-modal synchronization on a fixed-Hz or densest-stream clock
- 3D URDF trajectory playback and UMAP embeddings to spot outliers and redundant demos
- Per-modality statistics, the same normalization stats training consumes
- Dataset surgery: merge, clone, split, filter, and one-click ZIP export
State-of-the-art policies on managed GPUs.
Launch a training run in one call, from lightweight baselines to billion-parameter VLAs, on cloud GPUs Neuracore provisions, schedules, and fails over for you. Or bring your own model. No infra to run.
- 7 built-in algorithms: ACT, CNN-MLP, Diffusion Policy (+ flow-matching), π0, π0.5, GR00T N1.6
- Multi-GPU training, T4 → A100 80GB, 1–16 GPUs, multi-zone failover
- Bring your own model with real forward / backward / optimizer GPU validation
- Batch-size autotuning, checkpoint / resume, TensorBoard and cloud metrics
Deploy in one line and close the loop.
One Policy abstraction, three interchangeable backends. Feed a SynchronizedPoint in, get an action chunk out, the same structure you collected with.
- Three topologies: in-process (zero-latency), local server, or managed cloud endpoint
- Action chunking and hot checkpoint swaps without a restart
- Managed GPU endpoints with TTL auto-shutdown and per-hour metering
- Real-time closed-loop control: WebRTC live fusion, multi-rate sensors, ROS 2
The platform meets the tools you already run, on the hardware you already have.
ROS 2
Record from live topics and run policies as a node that subscribes and publishes actions.
Simulators
Isaac Lab, MuJoCo, and PyBullet rollouts land in the same dataset as real data.
Teleoperation rigs
Leader-follower arms, VR, space mice, and gloves through one ingest SDK.
Edge inference
Export to GPU or edge accelerators and confirm the control-loop rate before deploy.