Aigen has developed automated robots for farmers that scan images of crops to identify weeds and then remove the weeds. For the robots to work effectively, they need to process data and images quickly so they can identify and remove weeds impacting one plant and then immediately move on to the next plant. While the robots are designed to execute weed identification and removal on the edge of an IoT infrastructure, Aigen needed to implement a cloud environment for training the data models that allow the robots to improve their crop analysis and weed removal. To solve this challenge, Aigen turned to Avahi Technologies, which designed an Amazon Web Services (AWS) infrastructure to support the machine learning capabilities that Aigen required to train the data model. The solution speeds up data collection and produces a model with reduced errors that Aigen can iterate and then test in the field on the robots. Over time, continuously training the model will help the robots perform better in identifying and removing weeds.
About the Customer
To help build a future with no harmful chemicals used to grow food, Aigen is launching the world’s first AI-driven, network-connected, robotic service powered by the elements. The company’s robotic service reduces farmer workloads and their reliance on fossil fuels while also improving yield by autonomously navigating, analyzing, and weeding row crops. The vehicles offer a scalable, chemical-free way to grow healthier food—running 100% on renewable energy with solar panels doubling as sails to take advantage of the wind.
Customer Challenge: Accelerate the Speed of Data Model Training
Aigen has developed automated robots for farmers that scan images of crops to identify weeds and then remove the weeds. The robots allow farmers to reduce the cost of hiring manual labor and decrease their reliance on chemicals, some of which are now banned and others for which the cost is escalating. The robots also address the challenge of eradicating weeds that have evolved to become herbicide-resistant. For the robots to work effectively, they need to process data and images quickly so they can identify and remove weeds impacting one plant and then immediately move on to the next plant. In addition to analyzing plant scans, other variables the robots need to factor in to accurately identify weeds include the crop type, soil type, and the growing method used by each farmer. The robots must also consider the amount of sun, shade, and water that the plants are exposed to. While the robots are designed to execute weed identification and removal on the edge of an IoT infrastructure, Aigen needed to implement a cloud environment for training the data models that allow the robots to improve their crop analysis and weed removal. “Our vision-based system differentiates between weeds and crops and then mechanically strikes the weeds,” explains Hemdeep Dulthummon, Head of Strategy & Operations for Aigen. “For the robots to do this effectively, we need high confidence in the data that our training model collects.”
Speed is the key reason for training the data models in the cloud rather than on-premises. “There’s a small window of opportunity during the growing season to collect data and train the model,” says Dulthummon. “We wanted to create a compute resource foundation so that when our model fails, we fail fast. We then have more time to analyze the data and enhance the performance of the robots.”
Partner Solution: Avahi Builds Machine Learning Capabilities Into Serverless Architecture
To deploy the required cloud infrastructure to support the training models, Aigen first selected Amazon Web Services for its ability to provide a range of compute resources that are scalable, highly secure, and reliable—across server, storage, database, networking, data lake, analytics, machine learning, artificial intelligence, IoT, and security technologies. From there, Dulthummon found the ideal partner to design and deploy the AWS infrastructure—Avahi Technologies. Avahi specializes in helping start-ups like Aigen that need to collaborate with partners they can trust to secure intellectual property and quickly deploy an AWS infrastructure. Avahi also features a staff of experts with deep data modeling and AI knowledge. “Avahi impressed us with how they dived into scoping the project to understand our needs and define the deliverables we required,” Dulthummon says. “Given their cloud expertise, we were comfortable partnering with Avahi without considering anyone else.”
Dulthummon considers finding the right partner for this project right out of the gate a key benefit. “With a complex technology deployment like ours, we would have had to invest a lot of time in vetting multiple potential partners and bringing them up to speed so they could properly scope and quote the project,” says Dulthummon. “With the pace of our business, we did not have time to do this, but Avahi eliminated this issue by demonstrating they could get the job done on time. They even worked over a holiday to finish the proposal.”
To automate the flow of data and images from the Aigen robots connected to an IoT platform into the AWS environment, Avahi deployed an AWS Lambda serverless computing infrastructure. For coding, Avahi leveraged Terraform to create an infrastructure-as-code architecture, which streamlines updates to the environment and accelerates future deployments if Aigen needs to stand up another data modeling environment.
Key services of the architecture include AWS Machine Learning, which integrates with Labelbox, a third-party application for building and tuning data models. Another third-party app integration is Likely, which curates data for machine learning and helps Aigen select the right training data. The Avahi team took the time to understand both third-party applications to configure their APIs to interface easily with the AWS infrastructure as shown in the image below.
Other key services in the Aigen data modeling environment include Amazon API Gateway, AWS NAT Gateway Amazon Simple Storage Service (S3), Amazon Simple Queue Service (SQS), Amazon Elastic Container Service (ECS), and Amazon Relational Database Service. The architecture uses Amazon Cognito for identity and access management and Amazon Simple Notification Service (SNS) to automate message sending. Kubernetes CronJob allows Aigen to preschedule the data modeling tasks.
Results and Benefits: Enhanced Data Model Training Improves Robot Performance
Aigen collaborated with Avahi to set up the data modeling infrastructure so it speeds up data collection and produces a model with reduced errors that Aigen can iterate and then test in the field on the robots. Over time, continuously training the model will help the robots perform better in identifying and removing
weeds. The solution also segments Aigen’s golden dataset from the data model, a key attribute for machine learning, and it automates the uploading of farming data and images from the IoT platform on which the robots are networked. Avahi configured the AWS environment to categorize the data collected by the robots so Aigen can make sure the model has a good data set to work from.
“When we upload the data, we know where the data is and that it’s tagged properly,” says Dulthummon. “We then run the data through the machine learning process that includes sequencing and labeling so we can train the model and save it in the cloud for the robots to access.” The model also sorts the different types of crops and helps distinguish them from weeds in the scanned images. With the consistent training model that the AWS environment enables, Aigen can account for the different ways weeds grow depending on the variables. “As our robots weed, we measure our efficiency, collect the data, and improve our model,” Dulthummon
adds. “When training the model, we’re not always using the same dataset. We can pull data from different robots to find the optimal model. And with the AWS infrastructure, we know what data was used and what has it been trained on.”
With phase one of the project completed, Aigen is now discussing phase two with Avahi, which involves fine-tuning the data and image collection of the robots on the edge. By working with Avahi on phase one, Dulthummon knows the AWS infrastructure is set up properly to support phase two, and that the insights of the Avahi team helped avoid mistakes along the way, which ultimately accelerated the completion of the project. “We have solid technical experts who know what it takes to build a cloud infrastructure, but their focus is on our front-end application, and they don’t have the bandwidth to expand what they’re doing,” says Dulthummon. “By partnering with Avahi, we also reduced the risk of the project failing.”
In assessing the value of joining forces with a partner like Avahi, Dulthummon recommends that other software developers first evaluate how much knowledge they have in-house. It’s also key to know what your “genius zone” is. “Don’t try to be a genius at everything,” Dulthummon points out. “If you don’t have the right resources, just stick to your genius zone and partner with cloud experts like Avahi who take on these types of challenges every day. There’s a learning curve to everything and the opportunity cost is high. If you spend time on something you’re not expert in, it’s taking away from what you’re good at and which could be advancing your company.”
About the Partner: Avahi Technologies
Avahi Technologies is a cloud-native focused company and Amazon Web Services (AWS) partner with a team of cloud, data, and software engineering experts and experiences obtained through years of working within the cloud ecosystem. An extraordinary team of highly-certified Avahi experts excels in architecting and operating secure, automated, cloud-based solutions built on AWS. With a focus on becoming an extension to existing customer teams, Avahi offers exceptional service and works tirelessly to build the right solutions to solve business problems.