The second edition of Self-Racing Cars Event – an autonomous racing series. This year Udacity in partnership with PolySync and their OSCC project created team of talented engineers from Self-Driving Car Nanodegree.
Attendees
- Renovo Motors (Vehicle Platform for Autonomous/Connected/Electric)
- Comma.ai (Autonomous Vehicles)
- Point One Navigation (Spatial Localization for Autonomous Cars)
- Righthook (Vehicle simulation and testing)
- Revl (Action Camera)
- AutonomouStuff (Autonomous Hardware and Development Platform)
- Swift Navigation, Inc. (GPS for Autonomous Vehicles)
- PolySync (Autonomous Vehicle Software Platform)
- Xsens (IMU, AHRS, GNSS/INS)
- Udacity (Autonomous Vehicle Education)
- Compound Eye (Vision)
- NVIDIA (Self-Driving Car Platforms)
- Velodyne LIDAR (Sensing)
- CivilMaps (Cognition and Autonomous Mapping)
- “Right Turn Clyde” (Autonomous GoKart)
- Vector AI (Deep Learning Intelligence Vehicle)
- Sanborn (HD Maps for Autonomous Driving)
Be there – 360 experience!
Timeline
Feb 2 – Udacity announcement
Udacity announced they will create student team for Self-Racing Cars event in this Udacity’s medium post. I applied.
Feb 14 – Udacity Team is formed
From this moment our team started to work. First meeting video call, tasks splitting, timeline.
Simulation team starts to build Thunderhill West track simulator.
Models, models and more models…
First results with neural network. Net is predicting only steering angle for the car, speed is controlled outside. [Note after the event: Actually you want to drive with such net – you clearly know that it is working and trying to do something good]
First release! My first model which was working quite good even with higher speed.
After training few more models it was really good. I tried different augmentations and fine tuned the architecture.
Fully controlled by deep net:
- steering angle
- throttle
- brake
In first seconds net decides to start driving forward! [Note after the event: too slow, boring? Speed below 30mph is nice and comfortable when AI tries to kill you in the car!]
March 30 – Team meets the car
Finally! Lets start integration as fast as we can!
Steering angle is between -1 and 1, but -1 is right or left?!
There were lot of problems! Our trained models where not able to drive, we started collecting new data, to train on real car signals.
Apr 1-2 – Self-Racing Cars Event
The event started! A lot of fantastic cars, not only autonomous. We are ready for testing in production. 4 machines and one more in AWS are constantly training new models. Part of the team is gathering data on track, another part is processing data and the rest is training. When 5 or more new models are ready, we are testing them on track.
Enjoy the video from the race
Team Soulles Datasets
Datasets from the event are available at archive.org
More info
- Our Very Own Grand Challenge – medium post by Chris Gundling
- Team Soulless Youtube channel
- Udacity Self-Driving Car Engineer Nanodegree