Almond Brings Vision-Powered Robotics to Manufacturers

Teaching robots to handle complex tasks with vision AI

Industry
Manufacturing Technology
SOLUTIONS USED
-
AI-Assisted Labeling
-
Hosted Model Training
-
Cloud Inference Support
CONTACT US
Need help deploying computer vision? Talk to us.

Almond is a robotics company bringing AI-powered automation to manufacturing. They offer a complete solution – hardware, software, and service – to streamline the deployment of intelligent robotic arms for tasks like picking and machine tending. Unlike traditional robots that are limited to pre-programmed movements in controlled settings, Almond's AI-driven systems operate in varied environments and handle complex tasks.

“Around 90% of manufacturing robots leverage hard-coded movements,” said Shawn Patel, Co-Founder & CTO of Almond. “As a result, when dealing with more complicated tasks, manufacturers often resort to expensive facility alterations – for instance, installing vibrating mechanisms to move screws to specific spots so a pre-programmed robot can pick them. Now with artificial intelligence our robots can handle these complex tasks without needing to renovate your manufacturing line. The robot just drops in and starts working from day one.”

Mastering High-Mix Picking with AI

Almond has developed AI solutions for their robots to handle a diverse collection of tasks, from loading items into machinery to performing multi-point inspections of objects. One particularly challenging task was teaching their robots to pick and sort different shaped objects, such as pieces of sheet metal, that might be stacked atop each other.  

"These high-mix picking situations are perfect for vision AI," Patel explained. "Normally, if you have different shaped SKUs mixed together in a bin, it wouldn't be feasible for hard-coded robots to find and sort them. You'd need to run them in batches and then reprogram the robot after each batch. We wanted to develop a vision AI solution that would enable our robots to identify and sort the different SKUs – even when they were in a random pile."

This high-mix picking challenge led Almond to develop a multi-stage computer vision pipeline, combining foundational models along with purpose-built vision models they developed using Roboflow. "For humans, when you see a stack of items in a bin, understanding which one is on top and picking it up is simple,” Patel explained. “For a robot to do the same, we needed to combine several different vision AI models: one for identifying the different SKUs, one for determining which item was on the top of the stack, one for guiding the robot to the object's center, and so on."

When training the models for this multi-stage pipeline, the team at Almond experimented with different foundational models. For the model that needed to identify the center of each object, they trained a few different versions using YOLO and Roboflow’s RF-DETR. When comparing the early prototypes, their RF-DETR model significantly outperformed YOLO, achieving an 81% accuracy score compared to YOLO's 67%. This initial improvement in accuracy prompted Almond to focus on RF-DETR, ultimately training their model on over 2,000 images.

Streamlined Labeling and Training with Roboflow

Expanding the model dataset presented a new challenge: labeling thousands of images. "I thought we'd need to develop our own pipeline for automatically labeling images using the initial model," Patel said. "But Roboflow already had automatic labeling built into the platform. With their AI-assisted labeling tools, hosted training, model evaluation metrics, and – most importantly – highly accurate models like RF-DETR, Roboflow has simplified the process of developing state-of-the-art computer vision applications for our team."

The Result: Agile and Adaptable Robotic Automation

When deploying their robots, Almond leverages a mixture of computing infrastructure: simpler tasks run on local NVIDIA Jetson devices on the robots, while more complex computations are offloaded to Roboflow Inference in hosted cloud GPUs. This flexibility enables Almond to provide practical, AI-powered robotics solutions in real-world manufacturing environments.

Almond's innovative use of foundational and purpose-built vision models has resulted in highly adaptable robotics solutions for complex manufacturing tasks. This approach significantly reduces the need for costly facility modifications, allowing manufacturers to deploy robots that work immediately in various environments and solve previously intractable challenges.

About Almond

Almond is a robotics company focused on bringing AI-powered automation to manufacturers. They aim to streamline the experience of deploying AI-powered robots, such as automated arms for picking and machine tending, offering a complete solution including robotic hardware, software integration, and maintenance.

almondbot.com

Additional resources

Vision AI is transforming every industry. Let’s transform yours.

Speak with an AI expert

Our team will help you start solving business problems on the first call.

Ask us about

  • Solution architecting
  • Live demonstration
  • Pricing and specifications
  • Feasibility assessment
Message sent!
To pair you with the most relevant AI expert, please complete the next short form.
Oops! Something went wrong while submitting the form.
We will connect you with an AI subject matter expert on our team based on your answers.
What challenges would you like to solve with vision AI?
Where will you run vision AI?
Are you replacing a current solution with AI or will this be a new solution?
How many detections do you anticipate per month?
Describe the business problem you would like to solve.
Oops! Something went wrong while submitting the form.

Over 16,000 organizations build with Roboflow.

No items found.