TechCrunch Startup News Carbon Robotics built an AI model that detects and identifies plants
Feb 3, 2026
A look at an AI model that instantly recognizes plant species for laser weeding. Discussion of how farmers can mark new weeds live and avoid retraining. Details on the massive training set collected from farms worldwide. Notes on a software rollout, funding behind the tech, and plans to fine-tune with ongoing farm data.
AI Snips
Chapters
Transcript
Episode notes
Model Identifies Plants Instantly
- Carbon Robotics built a Large Plant Model (LPM) that recognizes plant species instantly across diverse appearances.
- The model generalizes from 150M+ labeled images so robots can identify new weeds without retraining.
Continuous Field-Fed Learning
- Carbon's robots feed field data back into LPM so the model continues improving as it operates.
- The company plans to fine-tune the model with ongoing data from machines across farms.
Past Retraining Took A Day
- Paul Mikasel explained that prior to LPM the company retrained models whenever a weed looked different or new appeared.
- Each retraining cycle took about 24 hours and interrupted real-time field response.