US company Bonsai Robotics has introduced Bonsai World, a simulation system that recreates real fields and working conditions in 3D so that autonomous farm machines can be trained and tested before they enter a new field, crop or task. The aim is to cut the on-site testing and adjustment that usually slows the arrival of a robot on a new farm.
Bonsai World is part of Bonsai Intelligence, the autonomy software that runs on the company's Amiga robotic platform and on machines from other manufacturers that it retrofits. It builds on more than 50 million real-world data samples collected across over 1 million acres (405,000 hectares), covering different crops, terrain, weather, lighting, machines and operations.
The system starts with satellite imagery of a farm. Google's Gemini vision-language model turns the flat image into a structured 3D environment in which a machine can simulate its routes and its interactions with the surroundings. The software can also generate realistic ground-level images and add situations robots meet in practice, such as dust, debris, animals, other vehicles and changing terrain, producing extra training data alongside what working machines already collect.
"Every environment we operate on expands what the system understands," said Tyler Niday, chief executive and cofounder of Bonsai Robotics. The company first focused on hard conditions such as unreliable GPS, weak connectivity and heavy dust, and Bonsai World is meant to carry that experience into new places without having to meet every situation in the field first. It was developed with Google Cloud and NVIDIA computing, with NVIDIA processors used to train the models.
Bonsai works in specialty-crop farming in the United States and Australia and is expanding into mining and defence. For growers, the promise is quicker deployment of autonomous machines in unfamiliar fields with less setup time. Future Farming notes, however, that Bonsai has not published independently verified results showing how much field testing the system saves or how closely performance in simulation matches work in a real field.
Source: Future Farming
Comments
(0)