San Jose farm-robot startup Bonsai Robotics adds simulation tool to train machines before deployment
Bonsai World turns satellite images of a farm or mine into a 3D practice ground, the company says.
San Jose-based Bonsai Robotics, which makes self-driving software for farm and off-road machines, has introduced Bonsai World, a tool that builds 3D simulations so its machines can practice on a site before they arrive, the company said in an Oct. 2 release.
Starting from satellite images of a farm, mine or similar site, the software builds a 3D model and adds conditions such as dust, animals, vehicles and changing terrain, the company said. Bonsai said this cuts the field testing needed when it enters new crops, sites or jobs. It uses Google's Gemini AI model and NVIDIA computing, according to the release. Bonsai joins a trend of companies creating and/or using synthetic data, made-up images that supplement real ones, to train machines more cheaply. NVIDIA builds virtual copies of real places it calls "digital twins" for the same purpose.
Bonsai said its models are trained on more than 50 million real-world samples gathered over more than 1 million acres. Its website says more than 400 of its machines are deployed for jobs such as spraying, mowing and weeding in almond and pistachio orchards, vineyards and strawberry fields.
"Every environment we operate on expands what the system understands," said Tyler Niday, the company's chief executive and co-founder.
Source: Bonsai Robotics, "Bonsai Robotics Unveils Bonsai World to Accelerate Physical AI Across Rugged Environments," Oct. 2, 2026; NVIDIA Glossary, "What Is Synthetic Data Generation (SDG)?"; NVIDIA Glossary, "What Is a Digital Twin?".