This repo contains the Jittor implementation of Bench2DriveZoo, which supports BEVFormer, UniAD , VAD in Bench2Drive. All models are student models of the world model RL teacher - Think2Drive.
本仓库是Bench2DriveZoo基于计图国产深度学习框架的实现,支持在闭环端到端自动驾驶测试基准Bench2Drive中运行。本项目在JAD基础上,将UniAD和VAD适配到CARLA仿真,使用基于世界模型的强化学习教师Think2Drive采集的数据进行模仿学习得到。
We also implement AD-MLP and TCP in Bench2Drive under Jittor. Use "git checkout tcp/admlp" to obtain their corresponding training and evaluation code.
我们也实现了Jiitor框架的AD-MLP和TCP在CARLA下的版本。请使用"git checkout tcp/admlp"切换到对应的分支。
- Installation (环境配置)
- Prepare Dataset (数据集准备)
- Convert Codes from nuScenes to Bench2DriveZoo-Jittor (迁移模型到本仓库)
-
Prepare your checkpoint, or download our pretrained models 准备好自己的checkpoint 或者 使用预训练好的模型。
-
Open the root directory, run: 在本目录下运行:
bash ./adzoo/uniad(vad)/uniad(vad)_jittor_eval.sh ./adzoo/uniad(vad)/configs/.../your_config.py /path/to/xxx.pth 1
e.g.:
bash ./adzoo/uniad/uniad_jittor_eval.sh ./adzoo/uniad/configs/stage2_e2e/base_e2e_b2d.py ./ckpts/uniad_base_b2d.pth 1
bash ./adzoo/vad/vad_jittor_eval.sh ./adzoo/vad/configs/VAD/VAD_base_e2e_b2d.py ./ckpts/vad_b2d_base.pth 1
- Follow Bench2Drive Official Repo to install CARLA and CARLA python egg.
- link the two agents
uniad_b2d_agent_jittor.py
andvad_b2d_agent_jittor
underleaderboard/team_code
folder of Bench2Drives - Modify the script "run_evaluation_debug.sh" to configure the team code agent, model config, and model checkpoint to run.
- Open the root directory of Bench2Drive, then
bash ./leaderboard/scripts/run_evaluation_debug.sh
Please consider citing the following papers if the project helps your research with the following BibTex:
如果您觉得本项目对您有帮助,请考虑引用:
@article{hu2020jittor,
title={Jittor: a novel deep learning framework with meta-operators and unified graph execution},
author={Hu, Shi-Min and Liang, Dun and Yang, Guo-Ye and Yang, Guo-Wei and Zhou, Wen-Yang},
journal={Science China Information Sciences},
volume={63},
number={222103},
pages={1--21},
year={2020}
}
@article{jia2024bench,
title={Bench2Drive: Towards Multi-Ability Benchmarking of Closed-Loop End-To-End Autonomous Driving},
author={Xiaosong Jia and Zhenjie Yang and Qifeng Li and Zhiyuan Zhang and Junchi Yan},
journal={arXiv preprint arXiv:2406.03877},
year={2024}
}
@inproceedings{li2024think,
title={Think2Drive: Efficient Reinforcement Learning by Thinking in Latent World Model for Quasi-Realistic Autonomous Driving (in CARLA-v2)},
author={Qifeng Li and Xiaosong Jia and Shaobo Wang and Junchi Yan},
booktitle={ECCV},
year={2024}
}