# Fei-Fei Li: Robots Need 'Spatial Intelligence' for Real-World Tasks _Fei-Fei Li and Yunus from Scenix discuss World Labs' vision for 'spatial intelligence' in AI and robotics, focusing on world models and the real-to-sim-to-real pipeline._ **Published:** 2026-07-28 **Source:** https://www.startuphub.ai/robotics/fei-fei-li-robots-need-spatial-intelligence-for-real-world-tasks --- In a recent appearance on The AI Podcast, AI pioneer and co-founder of [World Labs](/ai-news/investors-news/2026/ai-s-physical-future-beyond-the-model), Fei-Fei Li, discussed the future of artificial intelligence, focusing on the burgeoning field of "[spatial intelligence](/ai-news/ai-figures/2026/figure-fei-fei-li-recent-appearance-recap-2026-07-21)." Li, alongside Scenix co-founder Yunus, outlined their vision for building AI systems that can not only perceive and reason about spaces but also interact with them, both physically and virtually. Robots lack intelligenceDriver current AI systems struggle with understanding and interacting with physical and virtual spacesFrom the articleWorld Labs, with its foundation in spatial intelligence and world modeling, is poised to play a significant role in shaping the future of AI and robotics, aiming to make robots that truly "work" in the real world.needsSpatial IntelligenceContextAI's ability to generate, understand, reason with, and interact with environmentsFrom the article 5 mentionsLi explained that spatial intelligence is about creating AI that possesses a deep understanding of environments.World Labs VisionCoreFei-Fei Li and Yunus outline a future for AI systems with deep spatial understandingFrom the article 5 mentionsWhile World Labs' vision extends to various applications like VFX, gaming, and design, the physical interaction capabilities enabled by robotics are a significant focus.Robotics ApplicationEffectspatial intelligence is key for creating more capable and versatile real-world robotsFrom the article 7 mentionsWhile World Labs' vision extends to various applications like VFX, gaming, and design, the physical interaction capabilities enabled by robotics are a significant focus.usesReal-to-Sim-to-RealCorea crucial pipeline for training AI by bridging physical and simulated environmentsFrom the article 2 mentionsA key aspect of World Labs' approach is the development of a "real-to-sim-to-real" pipeline.World ModelsCoreAI systems that can predict future states and perform counterfactual reasoningFrom the article 9 mentionsWorld Labs is primarily focused on building "large world models," which are consistent and comprehensive representations of environments.Simulation PowerEffectleveraging synthetic data and data flywheels to accelerate AI developmentFrom the article 3 mentionsShe drew an analogy to human intelligence, noting that humans frequently use simulation in their minds for "counterfactual reasoning", playing out scenarios that haven't happened or cannot happen in reality.contributes toFuture of AIOutcomeachieving truly intelligent and interactive AI systems for complex tasksFrom the article 4 mentions"The ability to act within the physical space is one of the most exciting and profoundly important capabilities of the future AI world," Li said. ## Defining Spatial Intelligence Li explained that spatial intelligence is about creating AI that possesses a deep understanding of environments. "Spatial intelligence is about creating AI that has the ability to generate, understand, reason with, and interact with spaces, whether physical or virtual," she stated. This capability is seen as a crucial step towards achieving more capable and versatile robots. ## The 'Real-to-Sim-to-Real' Pipeline A key aspect of [World Labs](/ai-news/ai-news/2026/stripe-openrouter-atoms-infrastructure-week-july-2026)' approach is the development of a "real-to-sim-to-real" pipeline. Yunus elaborated on this concept, explaining its purpose: "We want to map the real environments into the digital world that has the best alignments with the real environments. By alignments, we mean that whatever happens in the digital world is also going to happen in the real environments." The goal is to replace the need for extensive real-world data collection and evaluation by leveraging scalable data generation in digital simulations. This allows for more efficient and safer training and testing of AI models, especially in robotics where data can be scarce and costly to acquire. ## World Models and Counterfactual Reasoning World Labs is primarily focused on building "[large world models](/ai-news/claude)," which are consistent and comprehensive representations of environments. Li emphasized the importance of consistency over space, time, and different interaction types. She drew an analogy to human intelligence, noting that humans frequently use simulation in their minds for "counterfactual reasoning", playing out scenarios that haven't happened or cannot happen in reality. This ability is critical for AI systems, especially robots, to learn and adapt effectively, as real-world data alone may not cover all possible situations. ## Robotics as a Key Application While World Labs' vision extends to various applications like VFX, gaming, and design, the physical interaction capabilities enabled by robotics are a significant focus. "The ability to act within the physical space is one of the most exciting and profoundly important capabilities of the future AI world," Li said. The recent acquisition of Scenix, a robotics company, is a testament to this commitment. Yunus highlighted that Scenix aims to solve key bottlenecks in general-purpose robotics, particularly around training and evaluation, through their real-to-sim-to-real pipeline. ## The Power of Simulation and Data Flywheels The discussion also touched upon the ongoing debate between simulation-based learning and real-world data collection. Both Li and Yunus stressed that these approaches are not mutually exclusive but rather complementary. Simulation provides reliability and efficiency through systematic randomization and coverage of the state space, while real-world data is essential for fine-tuning and validation. They believe that a "data flywheel" approach, where simulated and real-world data work in tandem, is the most effective path forward. ## The Future of Robotics and AI The conversation also delved into the future of robotics, with Li expressing a more measured outlook on humanoids, suggesting a preference for phased rollouts in more structured environments like warehouses before tackling complex unstructured settings. She noted that human bodies are optimized for general-purpose interaction in unstructured environments, a challenge that is exceptionally difficult to replicate in robots. The focus, therefore, is on specialized solutions and pragmatic approaches to achieve reliable and efficient robotic systems. World Labs, with its foundation in spatial intelligence and world modeling, is poised to play a significant role in shaping the future of AI and robotics, aiming to make robots that truly "work" in the real world. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.