What it’s like to build AI solutions with Carbon Robotics
In the world of agriculture, weeds are no small problem; they are a threat to farmers’ livelihoods.
Weeds compete with crops for important resources such as water, sunlight, and nutrients, limiting crop growth and reducing yields. And in some cases, weeds are so similar to crops that it becomes difficult for farmers to identify which plants to eradicate.
team of carbon robotics We want to help farmers meet this challenge more effectively. That’s why the company has developed a new large-scale plant model. This gives farmers precise control over the company’s LaserWeeder. LaserWeeder is a tool that uses AI, computer vision, and robotics to identify crops and weeds.
What does Carbon Robotics do?
carbon robotics Harness the power of computer vision, AI, and robotics to build autonomous agricultural equipment designed to eliminate chemicals, reduce costs, and increase yields.
Shosei Anegawa, a deep learning engineer who helped develop the model, said the new solution was trained on more than 150 million plant images and provides a new classification paradigm that greatly improves farmers’ ability to identify weeds that look nearly identical to crops.
For Anegawa, working with AI at Carbon Robotics is unique in the sense that off-the-shelf models are not suitable for the solutions he and his colleagues build. This means he can now build custom models that can detect small objects, distinguish between similar classes, and run in real time to weed with maximum efficiency, providing deep learning immersion, which he considers one of his favorite aspects of his job.
And the best part? Anegawa and his teammates see their ideas become products that positively impact farmers.
“I never get tired of walking behind the machine and watching the models and functions I designed kill weeds,” Anegawa said.
Below, Anegawa details the company’s new large-scale plant model and what it’s like to tackle AI with carbon robotics.
How carbon robotics leverages AI in agricultural technology
Learn how AI is used at Carbon Robotics. How does it relate to your product and product development?
We have developed a large-scale plant model trained on over 150 million plant images that powers LaserWeeder’s classification and targeting. Its main purpose is to give farmers precise control over the LaserWeeder’s classification and shooting behavior, allowing them to adapt to different field conditions and crops with a single model. We also employ other deep learning-based models for autonomous tractor kit tracking and targeting and recognition.
What is a large-scale plant model?
Carbon Robotics’ Large-Scale Plant Model is an AI model trained on over 150 million plant images that powers LaserWeeder’s classification and targeting. This allows farmers to identify weeds that closely resemble crops, allowing them to define plant behavior on the fly without retraining the model.
Can you tell us about any projects or milestones related to AI/ML and how it impacts carbon robotics?
The most impactful milestone so far has been the release of a large-scale plant model with an enhanced feature called Plant Profiles. Prior to LPM, our approach to weed detection and classification followed a standard object detection paradigm, training a model to locate and classify crops or weeds (one of the four weed categories we defined) depending on the crop the farmer was weeding. For example, farmers often don’t want to shoot certain plants, such as barley, that are planted to protect their crops. Alternatively, there may be specific weed species that both fall into the “grass” category but require very different laser exposure times to kill. Additionally, the concept of what a crop is varies by farmer and field, requiring a set of models for different crops.
To address this, large-scale plant models use a new approach that allows farmers to easily group and define species behaviors on the fly without retraining. This new classification paradigm has resulted in significant improvements in the treatment of weeds that have evolved to look almost identical to crops. This also gives you more options to grow most new crops and weed within minutes.
Carbon robotics approach to data management
Do you or your team have a unique approach to AI or data management?
I come from a research background, and one of the biggest differences about working at Carbon is that we have our own in-house labeling tools. This allows you to quickly add or change labeling strategies. This is not possible in research environments where datasets are often fixed. This opens up a whole new dimension of problem solving. You can also inject specific data into a model or change model behavior relatively quickly to solve a specific problem.
What it looks like for engineers to use AI at Carbon Robotics
What can AI/ML team members at Carbon Robotics do that others may not be able to do? Why should applicants interested in AI join your team?
One of the more interesting aspects of the problem we are solving is that off-the-shelf models often used for similar tasks, such as “You Only Look Once” algorithms for object detection or “unlabeled distillation” frameworks for self-supervised learning, either do not perform well or lack some characteristics that we need. This requires building custom models that can detect small objects, distinguish between highly similar classes, and run in real time to weed with maximum efficiency. This also applies to other parts of the machine that use AI, such as targeting and tracking. Using and designing custom models and pipelines to adapt to agriculture-specific challenges is one of my favorite parts of working with deep learning at Carbon.
How do carbon robotics engineers see the impact of their work on the real world?
What are the technical aspects of your work, individually or as a team?
What I’m most proud of is how close the loop is between ideas and machines that have a real impact on farmers’ fields. Whether it’s in their free time or when they’re dealing with more pressing issues, everyone is constantly working on experiments to solve specific problems that customers encounter, and when they reach the customer’s machines, they have a huge impact. We also go out into the field ourselves to experiment and test new features we design. This will give you more insight into the behavior of your model and help guide your work in a better direction. I never get tired of walking behind a machine and watching the models and features I designed work to kill weeds.
What does Carbon Robotics do?
Carbon Robotics combines computer vision, AI, and robotics to build autonomous agricultural equipment. Their main goal is to help farmers eliminate the need for chemical herbicides, reduce operating costs and increase crop yields.
What is a large-scale plant model?
Large-Scale Plant Model is an advanced AI model trained on over 150 million plant images. It powers Carbon Robotics’ LaserWeeder by providing a new classification system that allows the machine to identify weeds that look nearly identical to crops. A key feature of LPM is that farmers can group and define plant behavior on the fly without having to retrain the model.
How is Carbon Robotics using AI in agriculture?
Carbon Robotics is integrating AI into its agricultural technology in several important ways. The researchers are using macrophyte models to improve LaserWeeder’s ability to distinguish between crops and weeds, allowing the machine to accurately decide which plants to image with the laser and which to leave alone.
In addition to weed control, the company leverages deep learning models to recognize autonomous tractor kits, allowing the machines to move safely and effectively through fields.
