miniAutonomous
An   Artificial   Intelligence   Learning   Platform

miniAutonomous is an open-source, low-cost ($700) hardware and software platform created to develop and implement end-to-end steering and throttle control using deep learning!

The primary intent of the platform is to focus on the design and development of neural networks in the domain of end-to-end driving. The entire ecosystem was developed in native Python, uses minimal external packages, and contains no ROS components. It has been designed since its inception to be a clean, easy-to-use framework that has an on-vehicle graphical interface. The code base is a well-documented, highly modular and easily expandable software framework that can give those new to deep learning valuable experience in designing and deploying end-to-end networks.

Find Out More

Welcome!

We have released our on-vehicle software stack and the network trainer framework. Please stay tuned for pending updates to this portal and the respective repos.

Visit our GitHUB !

Stack Update

We are pushing out a series of updates across our stack which include:

  1. Creating a port of   trainer_ai   for PyTorch
  2. The vehicle control stack is being updated to Jetpack 4.6
  3. Replacing the Intel RealSense D425i with a Raspberry Pi CM2 sensor

As we continue to push our updates, we will update this section of the portal to reflect stack changes.

Our Mission

The goal is to provide an open-source platform that will allow those new to deep learning to design, train and deploy neural networks for autonomous tasks. To achieve this, we have designed an affordable hardware stack along with a software ecosystem that will allow the end-user to go through the entire network development cycle.

The Experience

We have created two software components that work in unison: engine_ai is the on-vehicle control framework that logs and labels data, facilitates the transfer of data to a server, and once a network is trained, allows for deployment of the network as an inference engine. trainer_ai is the server-side code base that leverages the data uploaded from the vehicle to train networks to complete the autonomous task.

Our objective is to provide all the basic elements needed to complete the network training cycle from data logging to network inference without any further coding requirement. The aspiration is that this serves as a base platform from which users can then design networks on their own to achieve a variety of objectives such as following-the-leader, outdoor track navigation, obstacle avoidance, etc. Although the user need not be proficient in Python or Tensroflow/Keras to get to functionality, the hope is that via the platform, users will gain insight into how the building block layers of neural networks work, what to consider when deploying them for edge computation, to get valuable experience with the software tools and to start forming basic intuition in network design.

In the days and weeks ahead, we will be adding videos for step-by-step assembly of the vehicle, Jupyter notebooks that provide insight into backpropagation, and a variety of other resources to help you build your own scaled vehicle.






Mini Autonomous Repositories


Get In Touch!

Have any questions, or would like to contribute, please contact us!

miniautonomous@gmail.com