TCP - Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline

Check origin repo from fork, here is for kin comparison

Docker and Dataset

build docker

git clone https://github.com/Kin-Zhang/TCP.git
cd TCP
docker build -t zhangkin/tcp .

download dataset

Download our dataset through GoogleDrive transfuser to server:

rsync -rvzc --progress -e 'ssh -p xx' --progress /home/kin/DATA_HDD/tcp_data/tcp_carla_data kinzhang@xx:~/data

run docker

run docker only for training, download their own dataset

docker run -it  --gpus all -v /dev/shm:/dev/shm -v /home/kin/DATA_HDD/tcp_data/tcp_carla_data:/home/kin/tcp/data --name kin_tcp zhangkin/tcp /bin/zsh

Training

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

python TCP/train.py --gpus NUM_OF_GPUS

Data Generation

First, launch the carla server,

cd CARLA_ROOT
./CarlaUE4.sh --world-port=2000 -opengl

Set the carla path, routes file, scenario file, and data path for data generation in leaderboard/scripts/data_collection.sh.

Start data collection

sh leaderboard/scripts/data_collection.sh

After the data collecting process, run tools/filter_data.py and tools/gen_data.py to filter out invalid data and pack the data for training.

Evaluation

First, launch the carla server,

cd CARLA_ROOT
./CarlaUE4.sh --world-port=2000 -opengl

Set the carla path, routes file, scenario file, model ckpt, and data path for evaluation in leaderboard/scripts/run_evaluation.sh.

Start the evaluation

sh leaderboard/scripts/run_evaluation.sh