Bonsai World turns satellite images into robot training environments
Bonsai Robotics introduced Bonsai World, a simulation application for machines working on farms, mines and other rugged sites. It turns satellite imagery into three-dimensional environments, then adds conditions machines may not have encountered in the field. The system extends the company’s existing autonomy platform with a way to train and evaluate machines before they move into unfamiliar surroundings.
Artificial Intelligence··Midday
Satellite views become virtual worksites
Bonsai Robotics, which develops autonomy software for machines in rugged environments, introduced Bonsai World on October 2. The application converts satellite imagery of farms or mines into structured three-dimensional simulations for training and evaluating autonomous machines. It is part of Bonsai Intelligence, the company’s existing environmental-intelligence layer, used in its Amiga platform and equipment adapted from other manufacturers.[1], [2]
Maps and ground views follow different steps
Google’s Gemini vision-language model interprets the satellite image and builds a structured map. Bonsai’s world model then generates ground-level views. Dust, debris, animals, vehicles and changing terrain can be added, allowing simulations to include conditions the fleet has not physically encountered. The company says the purpose is to prepare machines for unfamiliar environments before deployment.[1]
Simulation extends an existing autonomy platform
Training uses NVIDIA A100 processors on Google Cloud, while environment generation and inference use Blackwell server processors. Bonsai says its existing platform operates in specialty-crop agriculture in the United States and Australia and is expanding to other rugged industries. The new application extends that platform’s simulation tools. Reduced setup and field-tuning work remains a company expectation; independent field-performance measurements or public model weights did not accompany the announcement.[1]