Wayve

GAIA-3 is a generative world model from Wayve that transforms world modeling from visual synthesis into a foundation for autonomy evaluation. It generates lifelike, structured driving scenes with precise control, enabling safety-critical scenario generation, offline evaluation, and embodiment transfer. It combines realism of real-world data with control of simulation.

• •Generates lifelike driving scenes with structured and purposeful design
• •Enables controlled, realistic variations of real-world driving sequences
• •Supports safety-critical scenario generation (e.g., collisions, near-misses)
• •World-on-rails approach: alters ego trajectory while keeping other agents consistent
• •Embodiment transfer: re-renders same scene from new sensor configuration
• •Offline evaluation suites for repeatable and measurable testing
• •Consistency validation using LiDAR alignment

Detailed specs not disclosed.