TCP/ADMLP Bench2Drive

Checkpoint

• TCP
  • Hugging Face Link
  • Baidu Cloud Link
• ADMLP
  • Hugging Face Link
  • Baidu Cloud Link

Eval Json

• TCP-only-traj

Data Preprocess

• TCP

    # need set YOUR_Data_PATH
    python tools/gen_tcp_data.py

• ADMLP

    # need set YOUR_Data_PATH
    python tools/gen_admlp_data.py

Training

First, set the dataset path in TCP/config.py or ADMLP/config.py. Training:

    cd Bench2Drive-Zoo/
    # TCP
    export PYTHONPATH=$PYTHONPATH:PATH_TO_TCP
    python TCP/train.py --gpus NUM_OF_GPUS
    # or
    bash TCP/train.sh # need set your PATH_TO_TCP
    # ADMLP
    export PYTHONPATH=$PYTHONPATH:PATH_TO_ADMLP
    python ADMLP/train.py --gpus NUM_OF_GPUS
    # or
    bash ADMLP/train.sh # need set your PATH_TO_ADMLP

Open Loop Evaluation

    # TCP
    export PYTHONPATH=$PYTHONPATH:PATH_TO_TCP
    python TCP/test.py
    # ADMLP
    export PYTHONPATH=$PYTHONPATH:PATH_TO_ADMLP
    python ADMLP/test.py

Closed Loop Evaluation

Please follow these steps to evaluate TCP/ADMLP in Carla:

Preparations

• Install Bench2Drive from here.
• Follow this to install Carla.

Link this repo to Bench2Drive

    # Add your agent code
    cd Bench2Drive/leaderboard
    mkdir team_code
    cd Bench2Drive/leaderboard/team_code
    ln -s YOUR_TEAM_AGENT ./  # link your agent code

    cd Bench2Drive/
    ln -s Bench2DriveZoo/team_code/*  ./ # link entire repo to Bench2Drive

Run Evaluation

Follow this to use evaluation tools of Bench2Drive.