Machine Learning Car Project

Posted on

Here’s a machine learning car project I’ve mentioned to some of you over the last two meetings that I’m pursuing.  The parts finally arrived today! (picture below).  Several group members have expressed interest in participating, the more the merrier.  Some members also want to build their own car.  I’ve included some initial goals and the parts that I ordered below.

Initial Goals:
(a) Assemble and build the car robot and get it do all the usual out of the box things (e.g. control it via phone, line following, etc.)
(b) Add vision Machine Learning to it.  In particular, get a Tensorflow-based ConvNet working on the Rasberry PI so that the car can recognize/identify what it’s looking at.  Something similar to what Lukas did here:
(c) Add self-driving Machine Learning to it.  Have the robot learn to drive on its own (e.g. avoid obstacles around a track).  Was thinking of using Reinforcement Learning techniques inspired by OpenAI.  With any luck, try to port a model learned from a simulator onto the Pi.  For example, something similiar to this:
(d) The skies the limit, bring your ideas and creativity!


Parts I’m starting with:

If you use a different kit/parts, it might be best to stick with one that also uses a Rasberry PI 3 board so that we can share code, etc.  I chose this kit in particular because it claimed that they made all the source code available.  I purchased both components below on sale for ~$125.

(1) The Kuman Professional WIFI Smart Robot Car kit for Raspberry Pi, SM9 kit.  You can purchase from Amazon (currently unavailable) or from the manufacturer at .

(2)  Rasberry Pi 3 (e.g.