This article was last modified on 24 July 2017.
TensorFlow can become for machine learning what is OpenCV for computer vision and ROS as a robot operating system. Machine learning is getting more and more attractive to the public, and impressive results are not to be expected.
TensorFlow is an open-source library developed by Google and launched in 2015 for the general public. Ever since it was launched to be used by anyone who wants to work with artificial intelligence, the library broke all the records in terms of projects on GitHub.
Read now the best books to learn TensorFlow:
Returning to Google products, TensorFlow has a special place in the company’s projects, including Google Search, YouTube, Google Translate, Gmail, and more. Since the library is a very important resource of intelligence, in this article I made a list of examples using TensorFlow to do robotics applications. Furthermore, some of these examples are made by amateurs in the field of artificial intelligence.
Self-driving RC car using Raspberry Pi 3 and Tensorflow (added on 24.07.2017)
DeepTesla is a deep neural network designed to drive a car. The neural network is used in this project to recognize a path delimited by line.
The robot uses a simple webcam to capture images. The images are processed in real-time on a Raspberry Pi 3 at a rate of 30-50ms / frame.
The artificial intelligence module has been trained on a computer with a GTX 1060 graphics processor.
Wastesort based on opencv and tensorflow (added on 17.07.2017)
TensorFlow has the potential to change the recycling industry as we know it today.
Picture of a robot capable of sorting every piece of garbage with the same precision as a man. It is the next step towards recycling and artificial intelligence.
This robot is able to recognize at least four types of waste. To identify the category to which it belongs, the system uses TensorFlow and OpenCV.
Self driving Raspberry Pi with tensorflow and node-red (added on 26.06.2017)
TensorFlow can be used in several ways with Raspberry Pi. One way is through websockets.
In the movie below you can see how the robot changes its direction according to the images displayed on a computer screen.
- Object Recognizing Robot from $100 of Parts and TensorFlow
In this tutorial, Lukas Biewald used an ordinary 4WD platform for building a robot capable of detecting objects. The main components of the project are the camera, the Raspberry Pi board, and of course the TensorFlow library. The whole project cost around $100.
- Fully autonomous pi bot
The concept of self-driving cars has been around for years and is a hot subject in recent times. TensorFlow can be the basic library for this kind of project. In this example, Tawn Kramer uses a platform to track an interrupted line. The latest version of the project is a robot ready to run on its own.
- TensorFlow powered cucumber sorter by Makoto Koike
The industrial area may also be a field of use for TensorFlow. For example, sorting fruits and vegetables is a process that can be greatly simplified by using artificial intelligence. In the example below, TensorFlow is used to sort cucumbers.
- Object recognition robot
TensorFlow is very useful for recognizing objects. In this example, you can see how using the library, a camera, and a Raspberry Pi, you can detect up to 20,000 different objects.