search

AutoRally / autorally

Software for the AutoRally platform
http://autorally.github.io
727 stars 228 forks source link

Build Components and Prices #58

Open ksoltani opened 6 years ago

ksoltani commented 6 years ago

Hi, I was going through the excel sheet with all components and their prices. I have a few questions about the parts.

Some of the parts are quite expensive. For example the IMU, GPS and even the chassis. Were there any considerations given to cheaper options and if so it would help to know why cheaper options were not considered.

Additionally, it is not clear how cameras are being utilised in the control process. Are they used for obstacle avoidance?

If there were not cameras involved, would it be possible eliminate the onboard compute-box and use wireless commands?

bgoldfai commented 6 years ago

This issue should probably be over the autorally_platform_instructions repository, but I can answer here. As a side note, we're working on updated build instructions and parts list for an improved design and new sources for outdated parts. It should be available in the spring.

We chose the 1:5 chassis scale so that we could fit a mini-ITX form factor computer with an Nvidia GPU, and all the sensors we wanted, and it is the largest scale widely available in the hobby RC market. There are other open source platforms out there that are based on smaller RC cars: http://www.barc-project.com/ https://mit-racecar.github.io/

You could definitely go with cheaper IMU and GPS sensors for your build such as from Sparkfun, VectorNav, or Swift. If you go with a different sensor package we can't say how well our software components like the state estimator and controllers would work because we have not tested with another sensor package. We chose to go with the relatively expensive sensors out of the box to give us good sensor data without spending time on filtering and hacking interfaces.

Currently, the cameras are not used by any of the controllers, so you could eliminate them or replace them with webcams if you will not be doing much perception yourself.

All of our code currently runs onboard the compute box on the robot so even without cameras you can't remove the onboard compute box unless you can can create a high-reliability, high-bandwidth wireless link along with programs to feed the requisite sensor and command information back and forth, which we did not attempt to create due to the size and variety of our outdoor testing areas.