2. Background
• Machine Learning experience limited to one course in
Msc
• Experience mostly in web applications
• No experience in electronics
• Not looking to write a Deep Learning paper - just a
current interest in autonomous robotics and practical AI.
Example: how feasible is it to make an autonomous
vehicle or drone? Can I build a fleet of robocars that work
together on a task?
5. And probably has more lines of code than a
SpaceX rocket, with a computer hot enough to
toast your apartment in January!
6.
7. 1 $25 camera, 1 Raspberry Pi (A, B, B+), and a small
RC car, plus various <$15 electronic parts
2 hours hardware assembly
6 hours software setup
A few days to drive, train, tweak, and test
11. Machine Learning
• Same process as computational
photography on your smartphone
• Step 1 : capture image
• Step 2 : annotate image with motor & servo
output
• Step 3 & 4 on external PC: train model
• Step 5 back on Raspberry Pi : capture
image
• Step 6 on Pi : predict motor & servo output
• End-to-end approach. A supervised ML
model is trained to directly predict the motor
outputs from the original picture, without
passing through an intermediate
representation of vehicle speed or position
• Implementation of “Deep Learning for Self-
Driving Cars” by Nvidia team, Aug 2016
13. Next
• Add more sensors and more
actuators
• Use OpenCV/OpenMV
• Others on Github: RyanZotti,
Udacity, Apollo (biggest open
source project, a standard in
China) etc.
• Comma.AI
• Autonomous drones?
14. #Montreal
• Questions?
• The more, the merrier. Can’t have a race with just a
couple of people
• Community space? Track? Costs?
• Next Meetup : Meet & Work on your prototype. Foulab
has electronics and gear.