yuan16gcs / didi-starter

Creates a docker based starter kit for those working on the Didi challenge.

For now, it makes assumptions that you are running on ubuntu (16.04) and have an nvidia GPU installed and working.

Requirements

ahoy (v2)

To install on linux use: sudo wget -q https://github.com/ahoy-cli/ahoy/releases/download/2.0.0-beta1/ahoy-bin-uname -s-amd64 -O /usr/local/bin/ahoy && sudo chown $USER /usr/local/bin/ahoy && chmod +x /usr/local/bin/ahoy

try typing ahoy in your command line and you should see `ahoy - Creates a configurable cli app for running commands.'

nvidia-docker

Follow instructions here: https://github.com/NVIDIA/nvidia-docker#ubuntu-distributions and make sure you have your nvidia drivers and docker installed first.

docker and docker-compose

Best way is to use their repository for ubuntu: https://docs.docker.com/engine/installation/linux/ubuntu/#install-docker

Also, You'll want to setup a docker group with the right privilages so you don't have to type sudo every time. https://docs.docker.com/engine/installation/linux/linux-postinstall/

didi challenge data

Download the Didi-Training-Release-1.tar.gz dataset and extract it into the ./data subfolder. (don't commit anything in that folder into the repo please) You can also delete the data folder and replace it with a symlink if you need to, just don't commit that change.

Running things with ahoy

Running ahoy anywhere within this folder should show you the commands we've got working so far. The commands come from the .ahoy.yml file. Feel free to add more commands as needed.

Setup

You'll need to do these setup steps to start, and then any time you reboot.

ahoy x-up # Allows Xserver to accept connections from docker (use ahoy x-down to disable for security)
ahoy build # Builds the ros:nvidia container for use with GUI programs
ahoy network-up # Starts the docker 'didi-net' network so all the containers can talk to each other.
ahoy core-up # Starts the core services [master, cloud-node] and puts them in the network

Running Stuff

Here's a quick way to get started. Make sure you've put your data into the ./data folder though to make sure it works.

ahoy ros bash # A simple command that starts a new container with a bash shell. Good for testing things out.
ahoy rviz # Starts the rviz GUI using the default.rviz file settings. Should make it easy to get started.
ahoy rqt_bag # Starts the rqt_bag GUI. Useful for inspecting bag contents.
ahoy rosbag play data/approach_1.bag # will play the bag file and you should see point cloud and image data in rviz.

Issues encountered

need to run our custom Dockerfile to create ros:nvidia and then run that with ahoy desktop.

• Xserver rejects the request if you don't use xhost +local:root See http://wiki.ros.org/docker/Tutorials/GUI. Error shows as:

No protocol specified
rqt: cannot 

Use ahoy x-up to

• If you don't use nvidia-docker, I was getting an openGL error related to the swrast driver unless I used nvidia-docker. See openai/gym#509