Installation

This doc provides instructions to get started.

Install CARLA

• Install git lfs.
• Download this repo git clone --recurse-submodules git@github.com:dotchen/LAV.git
• Download and unzip CARLA 0.9.10.1

Install dependencies

• First, inside the repo, create a dedicated conda environment. Refer here if you do not have conda.

conda env create -f environment.yaml

• Inside the conda environment, install the CARLA PythonAPI easy_install [PATH TO CARLA EGG]. Refer to this link if you are confused at this step.
• Install PyTorch
• Install torch-scatter based on your CUDA and PyTorch versions.
• Setup wandb

Configure environment variables

Note: the following instructions only apply to Linux. For Windows follow these instructions instead.

Set the following environmental variables to your conda environment. Refer here for instructions to do so.

#!/bin/bash

export CARLA_ROOT=[LINK TO YOUR CARLA FOLDER]
export LEADERBOARD_ROOT=[LINK TO LAV REPO]/leaderboard
export SCENARIO_RUNNER_ROOT=[LINK TO LAV REPO]/scenario_runner
export PYTHONPATH="${CARLA_ROOT}/PythonAPI/carla/":"${SCENARIO_RUNNER_ROOT}":"${LEADERBOARD_ROOT}"
export TEAM_AGENT=[LINK TO LAV REPO]/team_code_v2/lav_agent.py
export TEAM_CONFIG=[LINK TO LAV REPO]/team_code_v2/config.yaml

export SCENARIOS=${LEADERBOARD_ROOT}/data/all_towns_traffic_scenarios_public.json
export REPETITIONS=1
export CHECKPOINT_ENDPOINT=results.json
export DEBUG_CHALLENGE=0
export CHALLENGE_TRACK_CODENAME=SENSORS

Now, you can treat this conda environment the dedicated one (by defauld named LAV-env).