Hello,
I am a student and I have to do an AI project in class.
I have started a project to make an automatic strawberry picker. I was able to do the AI with RoboFlow. But now I would like a microcontroller to make the connection between the AI and the servomotors that control the articulated arm. I was thinking about the Raspberry pi that would get the image and do the RoboFlow process either offline or with an API. Then the Raspberry Pi would send the information to an Arduino that would control the servo motors to place the arm in the milling cutters.
What do you advise me to do?
Hi Colin,
This sounds like a really cool project! If you need any more images, try searching “strawberry” in the Roboflow Universe search bar: Roboflow Universe, “strawberry” query results
I have a guide here from one of our interns from last summer on training a model with Roboflow Train to deploy to a Raspberry Pi board: https://blog.roboflow.com/rabbit-deterrence-system/
- The project’s GitHub repository: GitHub - roboflow-ai/rabbit-deterrence: Uses computer vision to deter rabbits from eating your vegetables
The guide includes sample code for sending images back to Roboflow to utilize Active Learning to quickly improve your model by programmatically sending images back to Roboflow to relabel and retrain. Active Learning will help to root out false detections and increase the confidence level of detections from your model.
Alternatively, there is the option to train a model with a model architecture like MobileNetSSDv2 and utilize the workflows from these blog posts and our Model Library for training and deployment:
-
MobileNetSSDv2 - Roboflow Model Library: MobileNetSSDv2 Tensorflow 1.5 Object Detection Model
-
Danger Monitoring for Cyclists with Raspberry Pi and Object Detection
-
How to Train a Custom Mobile Object Detection Model (with YOLOv4 Tiny and TensorFlow Lite)
-
Colab notebook to accompany the YOLOv4 Tiny and TensorFlow Lite guide
If you use the Colab notebooks, be sure to “Save a Copy to Drive” so the changes notebook are saved! Additionally, be sure to save your model weights.
What you’re doing is a lot like my weed sprayer with Raspberry Pi. I’ll point you to code when I get it up and perfected. At first it will be just the Y dimension. With Pi it is easy to control servos and relays!
Hi, do you get the solution? May I know what is your method?
We have full guides for deploying to Raspberry Pi, here:
Inference Server 1.0: Raspberry Pi (On Device) - Roboflow
Inference Server 2.0 (new) - use the ARM CPU container: