NeoteAI and Fudan push tactile sensing as a core scaling path for embodied AI
量子位 · wechat · 2026-07-26
- NeoteAI and Fudan University released a series of tactile-focused embodied AI reports: Foundation, VTLA, and TWAM.
- The core idea is that vision alone is not enough for real-world manipulation; tactile feedback is crucial for detecting contact, slip, overload, and edge collisions.
- NeoData provides a unified tactile data base: over 30,000 hours of visual-tactile interaction data, 1.4 million action clips, 3.3 billion time steps, 80 billion RGB frames, and 100 billion tactile frames.
- NeoForce learns a sensor-agnostic tactile representation to bridge different tactile hardware.
- VTLA predicts future tactile evolution up to 50 steps ahead and reportedly improves success rates on tasks like plug insertion and key extraction, while also learning from failure data.
- TWAM integrates video, tactile, and action prediction into a world model; it reports strong gains in simulation and on real robots.
- The team says the data and models will be open sourced, framing touch as a key scaling direction for embodied intelligence.
Related event: NeoteAI and Fudan Release Tactile Embodied AI Works(2 posts)→
More from Embodied
- Why robots don’t need human shapes, according to one AI hardware take — VraserX · 2026-07-26
- MicroFactory says its new robots can already assemble electronics — viktor_vrp · 2026-07-26
- Magic Leap exits its own AR headsets and becomes a waveguide supplier — 创业邦 · 2026-07-26
- CVPR 2025 paper learns 3D spatial audio from unlabeled video using camera ego-motion — 量子位 · 2026-07-26
- AI glasses are getting spatial workspaces with 3DOF tracking — Scobleizer · 2026-07-26
- REK Tokyo says Japan’s first humanoid robot show is a success — cixliv · 2026-07-26