Autonomy Hub Robotics Club
The Autonomy Hub Robotics Club (AHRC) is formed to make and race pro-level autonomous cars on a budget. That means that they’re smaller than regular cars (from go-kart size, down to 1/16th scale) and can be used indoors.
But just because they’re small and cheap doesn’t mean that we cannot run real autonomous car software on them. Rather than doing all processing on-board, these robotic cars tend to transmit the data from their on-board sensors (cameras, sonar, lidar, radar, GPS or whatever else we have) via Wifi to a laptop that runs pro-grade AI and robotics software, including TensorFlow, ROS and the rest of the Udacity Self-Driving Car nanodegree toolchain.
AHRC will gather regularly to hack and race its robotics cards at the Autonomy Hub headquarters in Columbus, Ohio. With plans for an indoor and outdoor track, AHub is working on setting up its first meetup August 19, 2017.
You can make almost any RC car self driving using the donkey library, but we recommend you build the Donkey2 which is a tested hardware and software setup. You can buy all the parts for ~$200 on Amazon and it takes -2 hours to assemble.
- Build your own toy car that can drive it's self.
- Drive your car with your phone.
- Record images, steering angles & throttles.
- Train neural net pilots to drive your car on different tracks.
- Race your car in a DIY Robocars race.
- Cars must start with a single binary interaction. This could be a button can be on the car, on a controller, a key on a keyboard or equivalent. No other intervention can happen until after the race is over otherwise the car gets a DNF. An emergency stop button is recommended but not required. Deadman’s switch is also acceptable where button is pushed and held for the duration of the race.
- There are no rules governing where the computing needs to take place. Cars may have onboard computing, trackside computing or leverage remote or cloud resources.
- GPS and other similar positioning systems are not allowed except for specified outdoor races.
There are detailed steps and learning to build out this vehicle, but it is manageable and adaptable to larger vehicles one it is tested out on this size.
Ahub does not have a 3D printer and will need one or can order parts from Shapeways or 3Dhubs.
Type of Tracks
There are two tracks, one with road-like white lines and yellow center line for ~1/10th scale autonomous cars, and another one for smaller cars around. Rules and descriptions for both are below:
- While it is called a 1/10 track, in fact, any size car smaller than 1/8 can be used, however cars that are much smaller than 1/10 will be at a significant disadvantage in time trials and can get damaged
- To complete the track the car must go around all cones at the corners and not run into them. Failure to go around the cones will trigger a DNF
- Out of bounds is defined as cars that go way outside of the track to the point where they will never get back on track.
- Course measurements are roughly accurate, but your bot should be able to handle variation, dirt, variable lighting, ect
- Black (optional, only if want to really make sure a car never fails to detect the edge)
- Lighting is important for computer vision. We need to lay out the track in a place where we can get consistent lighting. Good artificial lighting is better than sunlight that comes and goesIndoor Lanes:
The race will include 3 laps timed with both the first lap time and the total time recorded. The first prize will go to the fastest 3 lap time, unless nobody finishes 3 laps then the prize goes to the fastest first lap time.
Watch a virtual warehouse simulation race below.
Wheel to wheel races and redo races will be exhibition.
Each car that attends will be recorded as
- DNR – Did Not Race
- DNF – Did Not Finish
- Time – This is the time of the race.
Attendance will ultimately be important for governance etc.