Skip to content

Latest commit

 

History

History
112 lines (97 loc) · 3.21 KB

stack5_carla_openpilot.md

File metadata and controls

112 lines (97 loc) · 3.21 KB
Raw

Setup

Requirements

  • Monitor (i.e., due to the limitation of OpenPilot, the simulation can only run on a machine with a monitor/virtual monitor)
  • OS: Ubuntu 20.04
  • CPU: at least 6 cores
  • GPU: at least 6GB memory
  • Openpilot 0.8.5 (customized)
  • Carla 0.9.11

Directory Structure

~(home folder)

├── openpilot
├── Documents
│   ├── self-driving-cars (created by the user manually)
│   │   ├── ADFuzz
│   │   ├── carla_0911_rss

Note: one can create link for these folders at these paths if one cannot put them in these paths.

Install OpenPilot 0.8.5 (customized)

In ~,

git clone https://github.com/AIasd/openpilot.git

In ~/openpilot,

./tools/ubuntu_setup.sh

In ~/openpilot, compile Openpilot

scons -j $(nproc)

Common Python Path Issue

Make sure the python path is set up correctly through pyenv, in particular, run

which python

One should see the following:

~/.pyenv/shims/python

Otherwise, one needs to follow the displayed instructions after running

pyenv init

Common Compilation Issue

clang 10 is needed. To install it, run

sudo apt install clang

Common OpenCL Issue

Your environment needs to support opencl 2.0+ in order to run scons successfully (when using clinfo, it must show something like "your OpenCL library only supports OpenCL <2.0+>")

Install Carla 0.9.11

In ~/Documents/self-driving-cars,

curl -O https://carla-releases.s3.eu-west-3.amazonaws.com/Linux/CARLA_0.9.11_RSS.tar.gz
mkdir carla_0911_rss
tar -xvzf CARLA_0.9.11_RSS.tar.gz -C carla_0911_rss

In ~/Documents/self-driving-cars/carla_0911_rss/PythonAPI/carla/dist,

easy_install carla-0.9.11-py3.7-linux-x86_64.egg

Install additional maps

In ~/Documents/self-driving-cars/carla_0911_rss,

curl -O https://carla-releases.s3.eu-west-3.amazonaws.com/Linux/AdditionalMaps_0.9.11.tar.gz
mv AdditionalMaps_0.9.11.tar.gz Import/

and then run

./ImportAssets.sh

Run Fuzzing

python ga_fuzzing.py --simulator carla_op --n_gen 10 --pop_size 50 --algorithm_name nsga2 --has_run_num 500 --episode_max_time 200 --only_run_unique_cases 0 --objective_weights 1 0 0 0 -1 -2 0 -m op --route_type 'Town06_Opt_forward'

Rerun previous simulations

In ~/openpilot/tools/sim/op_script,

python rerun_carla_op.py -p <path-to-the-parent-folder-consisting-of-single-simulation-runs-indexed-by-numbers>

Rerun scenarios using the best sensor prediction

Move all the subfolders indexed of previously run simulation results in ~/Docuements/self-driving-cars/2020_CARLA_challenge/run_results_op to openpilot/tools/sim/op_script/rerun_folder, then in openpilot/tools/sim/op_script,

python rerun_carla_op.py -p rerun_folder -m2 best_sensor -w 2.5

Check the number of unique violations in terms of coverage

In openpilot/tools/sim/op_script,

python trajectory_analysis.py -p <parent folder> -f <fusion folder>

Analyze fusion errors in terms of objectives

Rename the original folder as "original" and the rerun_folder as "rerun_2.5_best_sensor". Put them into the same parent folder. In openpilot/tools/sim/op_script,

python analyze_fusion_errors.py -p <parent_folder>