North Star
A holistic intelligence platform for robotics, centered on touch.
Find tactile and vision+tactile datasets, download or fine-tune VLA/policy models, run open benchmarks and evals, and simulate contact-rich tasks — in one place. The bet: touch closes the failure gap vision-only systems hit on slip-prone, transparent, and deformable objects, and a shared data → model → eval → sim loop is what turns that into a usable platform.
Platform map
Three areas · 16 capabilities
Search
Find, vet, and acquire tactile & vision+touch data.
Datasets
Browse, filter, and preview datasets & episodes by sensor, format, license, task, object, robot, environment, and outcome.
Datasets
Every open-source tactile dataset, scraped daily from Hugging Face, GitHub, and Zenodo — searchable by sensor, modality, format, task, and license.
Datasets
Search captured video sessions and play their synchronized tactile hand-pressure data.
Toolkit
Capture synchronized vision + tactile glove data in-browser (Touchtronix and beyond) — live upload to the catalog plus a local copy.
Toolkit
Ingest capture videos into the catalog — paste a Google Drive folder link and every video is uploaded to the backend server-side.
Data Services
Request-a-collection service and licensable tactile data adaptors — request a spec, get a capture plan.
Toolkit
Multimodal sync (camera + tactile + F/T on one timeline), custom sensor readouts, and dataset quality/coverage checks before training.
Intelligence
Train, evaluate, and transfer policies across sensing setups.
Models
A hub of open-source and Monty-fine-tuned models behind one interface for both vision-only and tactile-fusion architectures.
Models
Training runs across datasets and sensing setups — pick architecture + sensing setup + datasets and launch.
Benchmarks
Compare runs by success rate, trial count, and 95% CIs; breakdown curves and failure-replay into the viewer.
Benchmarks
Public, reproducible benchmarks — ALOHA sim, LIBERO, DexBench, ContactWorld, and an open-touch benchmark with leaderboards.
Models
Cross-sensor / cross-robot transfer-score heatmaps and shared representations (contact-edge encoder, incipient-slip head).
RL Environments
Compose any dataset + any model against a specific task for reinforcement learning and policy evaluation.
Studio
Design sensor surfaces and simulate contact-rich touch.
Studio
Design tactile sensor surfaces end-to-end: shape → application → sensor placement → deployment, with 3D preview, live data, and CAD/PDF export.
Studio
The full sensor-surface design wizard on a pre-seeded example — no login required.
Simulation Platform
Physics sandbox: real-time sensor deformation modeling, geometry tuning, and force/ADC prediction.