PodBrowser
a16z

World Models, Robotics, and the Future of 3D AI

Sunday, 13 September 2026 · 2 min read · Listen to the episode ↗

In this episode, Justin Johnson introduces Atlas, a revolutionary multimodal world model that generates image and video frames with precise 3D reconstruction capabilities. The discussion highlights Atlas's potential applications across industries like entertainment and robotics, emphasizing its ability to create immersive environments and enhance robotic interactions.

The episode centers on the concept of world models, with Justin Johnson introducing Atlas as a groundbreaking multimodal world model. Atlas is capable of generating image and video frames with pixel-perfect camera control while reconstructing them in 3D, marking a significant advancement in the field.

Atlas is designed to create new worlds, reconstruct existing spaces, and simulate environments across various industries, including entertainment, virtual reality, construction, and robotics. Johnson anticipates that Atlas will facilitate the creation of 3D Gaussian splat scenes for game development, although he acknowledges that the internal tooling for steering Atlas will need further refinement to fully leverage its capabilities.

Johnson emphasizes that Atlas differs fundamentally from Marble, as it represents a completely rebuilt stack of models while maintaining the core principle of multimodal capabilities. This allows for spatial control by grounding reference images in 3D space, enabling precise camera steering during video generation.

The episode also explores future advancements, such as the potential for real-time pixel streaming from models via powerful server farms. However, Johnson cautions that the shift to fully AI-native workflows may take longer than expected due to entrenched industry practices. Simulation is highlighted as a crucial tool for evaluating robotics in their operational environments, with pre-trained robotics foundation models being fine-tuned in simulations that closely match their end applications.

Atlas's ability to reconstruct spaces from casual videos or images enhances interactions in robotics. The discussion touches on OpenAI's transition from World Models to consumer products, illustrating a potential innovator's dilemma. Despite this, Johnson asserts that World Models can serve as a versatile platform for various applications, unlocking substantial value.

Predictions indicate that mass consumer applications of World Models will likely emerge in virtual reality, augmented reality, and video games, with a potential shift towards vibe coding in game development. The aim is to make World Models quickly deployable for public use, promoting interoperability with other tools and empowering users to create innovative experiences.

The episode concludes with the assertion that while new technologies in creative fields will enhance user experiences, they will not replace the necessity for human creativity. Atlas is particularly noted for its capability to perform bullet time shots with synchronized captures, and its native camera control is recognized as a valuable input type in creative workflows.

This summary was generated from the episode transcript and can contain mistakes.