Hello Robotaxi launches driving world model HelloWorld

Self-driving R&D has long faced a data problem: rare scenarios and extreme conditions are hard to collect, which slows down training and testing of driving strategies.

Hello Robotaxi has announced HelloWorld, a generative driving world model. It builds driving data on demand and tests strategies interactively, combining scene control, continuous generation, multi-sensor observation, and fast inference.

The model accepts natural language requests or direct edits to vehicles, pedestrians, and motion paths to create multi-view videos and lidar data. Developers can add crossing pedestrians, change a car's route, or switch between rain, snow, day, and night to cover rare cases.

Scene descriptions like construction zones, heavy traffic, and pedestrians crossing, plus weather and lighting settings, can all guide the generation, offering a flexible way to build test scenes.