Grounded API launches with SOTA hand-tracking (<1cm) and SLAM, turning ego data into robot training data
pabbeel · x · 2026-09-17
Robotics data startup Grounded has launched its Grounded API, positioning itself as infrastructure for robot training data:
- SOTA on hand-tracking benchmarks (<1 cm error)
- SOTA on SLAM benchmarks
- Converts in-the-wild egocentric data into enriched training data in minutes
- Integrates with Hugging Face LeRobot and rerun
- Built for BitRobot Network's RoboCap suite
NYU professor Lerrel Pinto amplified the launch, arguing that while human data is heading toward abundance, turning it into an abundance of robots remains the open question. A technical report accompanies the release.
Related event: Grounded Launches API With SOTA Hand Tracking and SLAM(3 posts)→
More from Embodied
- Open-source Multi-Agent CAD generates editable CAD assemblies for ~$0.15 per part, 1/13 the cost of CAD Skills — burhop · 2026-09-17
- Princeton and Basis recruit a postdoc for MARA, robotic agents learning code-based world models — tomssilver · 2026-09-17
- After 'Perv Glasses' Backlash, Meta Preps A Camera-Free Pair Of Smart Glasses — TechCrunch AI · 2026-09-17
- Arcturus Vision Sync launches: move Steam Frame spatial videos to Windows & Mac — kscottz · 2026-09-17
- Dahlia: open-source Python library lets AI agents drive SOLIDWORKS for CAD work — s4chinraja · 2026-09-17
- China's UBTECH Opens Plant Making a Humanoid Robot Every 10 Minutes — yogthos · 2026-09-17