Home

Awesome

CARLA-Roach

<p align="center"> <img width="50%" src="doc/teaser.png"> </p> <!-- <img src="doc/teaser.png" align="center" width="66%"> -->

This is the official code release of the paper <br> End-to-End Urban Driving by Imitating a Reinforcement Learning Coach <br> by Zhejun Zhang, Alexander Liniger, Dengxin Dai, Fisher Yu and Luc van Gool, published at ICCV 2021. <br>

It contains the code for benchmark, off-policy data collection, on-policy data collection, RL training and IL training with DAGGER. It also contains trained models of RL experts and IL agents. The supplementary videos can be found at the paper's homepage.

The Offline Leaderboard

The "Leaderboard" we evaluated on is an offline version of the CARLA Leaderboard. As further detailed in the paper, the offline Leaderboard has the following setup

One can use the offline Leaderboard if a thorough study on the generalization ability of the method is desired.

Pros and Cons of the Online Leaderboard:

(+) All methods are evaluated under exactly the same condition.

(+) No need to re-evaluate other methods.

(-) No restriction on how the method is trained and how the training data is collected.

Pros and Cons of the Offline Leaderboard:

(+) Strictly prescribes both the training and testing environment.

(+) Full control and observation over the benchmark.

(-) You will have to re-evaluate other methods, if any setup of the benchmark has changed, for example CARLA version and etc.

Installation

Please refer to INSTALL.md for installation. We use AWS EC2, but you can also install and run all experiments on your computer or cluster.

Quick Start: Collect an expert dataset using Roach

Roach is an end-to-end trained agent that drives better and more naturally than hand-crafted CARLA experts. To collect a dataset from Roach, use run/data_collect_bc.sh and modify the following arguments:

Benchmark

To benchmark checkpoints, use run/benchmark.sh and modify the arguments to select different settings. We recommend g4dn.xlarge with 50 GB free disk space for video recording. Use screen if you want to run it in the background

screen -L -Logfile ~/screen.log -d -m run/benchmark.sh

Trained Models

The trained models are hosted here on W&B. Given the corresponding W&B run path, our code will automatically download and load the checkpoint with the configuration yaml file.

The following checkpoints are used to produce the results reported in our paper.

Available Test Suites

Set argument test_suites to one of the following.

Collect Datasets

We recommend g4dn.xlarge for dataset collecting. Make sure you have enough disk space attached to the instance.

Collect Off-Policy Datasets

To collect off-policy datasets, use run/data_collect_bc.sh and modify the arguments to select different settings. You can use Roach (given a checkpoint) or the Autopilot to collect off-policy datasets. In our paper, before the DAGGER training the IL agents are initialized via behavior cloning (BC) using an off-policy dataset collected in this way.

Some arguments you may want to modify:

Collect On-Policy Datasets

To collect on-policy datasets, use run/data_collect_dagger.sh and modify the arguments to select different settings. You can use Roach or the Autopilot to label on-policy (DAGGER) datasets generated by an IL agent (given a checkpoint). This is done by running the data_collect.py using an IL agent as the driver, and Roach/Autopilot as the coach. So the expert supervisions are generated and recorded on the fly.

Most things are the same as collecting off-policy BC datasets. Here are some changes:

Train RL Experts

To train RL experts, use run/train_rl.sh and modify the arguments to select different settings. We recommend to use g4dn.4xlarge for training the RL experts, you will need around 50 GB free disk space for videos and checkpoints. We train RL experts on CARLA 0.9.10.1 because 0.9.11 crashes more often for unknown reasons.

Train IL Agents

To train IL agents, use run/train_il.sh and modify the arguments to select different settings. Training IL agents does not require CARLA and it's a GPU-heavy task. Therefore, we recommend to use AWS p-instances or your cluster to run the IL training. Our implementation follows DA-RB (paper, repo), which trains a CILRS (paper, repo) agent using DAGGER.

The training starts with training the basic CILRS via behavior cloning using an off-policy dataset.

  1. Collect off-policy DAGGER dataset.
  2. Train the IL model.
  3. Benchmark the trained model.

Then repeat the following DAGGER steps until the model achieves decent results.

  1. Collect on-policy DAGGER dataset.
  2. Train the IL model.
  3. Benchmark the trained model.

For the BC training,the following arguments have to be set.

During the DAGGER training, a trained IL agent will be loaded and you cannot change the configuration any more. You will have to set

Citation

Please cite our work if you found it useful:

@inproceedings{zhang2021roach,
  title = {End-to-End Urban Driving by Imitating a Reinforcement Learning Coach},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  author = {Zhang, Zhejun and Liniger, Alexander and Dai, Dengxin and Yu, Fisher and Van Gool, Luc},
  year = {2021},
}

License

This software is released under a CC-BY-NC 4.0 license, which allows personal and research use only. For a commercial license, please contact the authors. You can view a license summary here.

Portions of source code taken from external sources are annotated with links to original files and their corresponding licenses.

Acknowledgements

This work was supported by Toyota Motor Europe and was carried out at the TRACE Lab at ETH Zurich (Toyota Research on Automated Cars in Europe - Zurich).