Fei-Fei Li on Spatial Intelligence: WorldLabs Acquires SceniX to Build Robotic Digital Training Grounds
量子位 · wechat · 2026-08-06
Fei-Fei Li sat down with a16z to deeply analyze WorldLabs' layout in spatial intelligence and robotics, announcing the acquisition of simulation company SceniX.
Core Visions & Strategy
- Spatial Intelligence as the Next Frontier: Enabling AI to generate, understand, reason, and interact in real or virtual spaces. Robotics is the first touchstone for landing spatial intelligence.
- Acquisition Logic: SceniX initially started as a customer of WorldLabs' generative model, Marble. The teams found high synergy, leading to the merger. SceniX co-founder Yunzhu Li was a postdoc under Fei-Fei Li.
Solving the Robotic Data Bottleneck
- Unleashing Scaling Law: Robotics severely lacks real-world training and evaluation data. Simulation environments are essential to unleash the power of scaling laws.
- Real-to-Sim-to-Real Pipeline: SceniX maps real environments into highly aligned digital worlds, replacing expensive and dangerous real-world evaluations with controllable, efficient digital twins.
- Counterfactual Reasoning: Li emphasized that simulation handles counterfactual reasoning that real data cannot (similar to Waymo's autonomous driving simulations), boosting both reliability and efficiency.
Technical Route Analysis
- 3D & Simulation vs. Pure Video Models: Pure video prediction models often suffer from physical inconsistencies (e.g., objects disappearing). Infrastructure combining 3D reconstruction and simulation provides initial momentum for the data flywheel.
- Models Need Not Be Perfect: Simulation must capture the essential structure of the problem, combining physics and learning to gradually transition from simulation-driven to real-world executable policy models.
More from Embodied
- EgoHumanoid: Training Humanoid Robots via Egocentric Human Videos — micoolcho · 2026-08-06
- UMD Introduces HumanEgo: Training Robots via Human Demonstration Videos — furongh · 2026-08-06
- Open-Source DIY Translator Built on Raspberry Pi and Gemma — osanseviero · 2026-08-06
- Minimax video model runs locally: 4080 generates 1280x720 in ~30 mins — cocktailpeanut · 2026-08-06
- Developer working on reading experience for book scanner — aaraalto · 2026-08-06
- Three New Low-Cost Robots Unveiled Starting at $1,688 — Scobleizer · 2026-08-06