# RESCO
![Alt text](maps.png?raw=true "Provided SUMO scenarios")
Source code implementing the Reinforcement Learning Benchmarks for Traffic Signal Control (RESCO).
The benchmark uses the Simulation for Urban Mobility (SUMO), which must be [installed separately](https://sumo.dlr.de/docs/Installing/index.html). SUMO_HOME environment variable must be set, this is done automatically on the install of Sumo on Windows and Ubuntu. SUMO 1.9.0 and 1.9.1 have been tested.
On Ubuntu the speed of the simulation may be greatly increased by using libsumo. Set the environment variable LIBSUMO_AS_TRACI to any value and give main.py --libsumo True. Note that this can not be used with multi-threading.
Python 3.7.4 is required for tensorflow -used by the MA2C and FMA2C implementation.
agent_config defines parameters for the available agents. An agent is specified by the --agent argument to main.
map_config specifies the SUMO scenario parameters, road network, and demand files.
mdp_config supplies constants to state and reward functions (e.g. for normalization)
signal_config defines each signal of each SUMO scenario. Valid green phases are determined from the road network TLSLogic, yellow signals are inserted as required. phase_pairs gives the directional index of phase combinations following the order defined in TLSLogic. valid_acts provides a translation table for shared controllers with varying action definitions across multiple signals. For each signal inbound lanes are given by the direction of traffic. Finally, each signal defines which signals are downstream for the purposes of coordination (neighbors, pressure, etc.)
An example command to train IDQN on the Ingolstadt region scenario is:
`python main.py --agent IDQN --map ingolstadt21`
SUMO scenarios are supplied in the environments directory. All scenarios are distributed under their original licenses. Information on the Cologne scenario can be found on (https://sumo.dlr.de/docs/Data/Scenarios/TAPASCologne.html). Information on Ingolstadt scenarios can be found at (https://github.com/silaslobo/InTAS). For more scenarios please see (https://sumo.dlr.de/docs/Data/Scenarios.html)
Below the benchmark performance for baselines (Fixed Time, Greedy, Max Pressure) and learning algorithms (IDQN, IPPO, MPLight, Extended MPLight (MPLight*), FMA2C) are given.
![Alt text](delays.png?raw=true "Benchmark learning curves")
# Citing RESCO
This project was used in [Reinforcement Learning Benchmarks for Traffic Signal Control](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/f0935e4cd5920aa6c7c996a5ee53a70f-Abstract-round1.html). If you use RESCO in your work, please include a citation:
```
@inproceedings{ault2021reinforcement,
title={Reinforcement Learning Benchmarks for Traffic Signal Control},
author={James Ault and Guni Sharon},
booktitle={Proceedings of the Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS 2021) Datasets and Benchmarks Track},
month={December},
year={2021}
}
```
# EPyMARL
RESCO has been updated to be compatible with the [EPyMARL](https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/a8baa56554f96369ab93e4f3bb068c22-Abstract-round1.html) benchmark for cooperative RL algorithms. Some modifications within EPyMARL are required currently, available [here](https://github.com/Pi-Star-Lab/epymarl_resco). Clone the modified repository and execute EPyMARL algorithms against the RESCO benchmark using EPyMARL's main.py:
```main.py --config=qmix --env-config=gymma with env_args.time_limit=1 env_args.key=resco_benchmark:cologne3-qmix-v1```
没有合适的资源?快使用搜索试试~ 我知道了~
交通信号控制(RESCO)的强化学习基准_Python_下载.zip
共69个文件
py:29个
xml:14个
sumocfg:8个
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 101 浏览量
2023-04-27
10:54:15
上传
评论
收藏 32.26MB ZIP 举报
温馨提示
交通信号控制(RESCO)的强化学习基准_Python_下载.zip
资源推荐
资源详情
资源评论
收起资源包目录
交通信号控制(RESCO)的强化学习基准_Python_下载.zip (69个子文件)
RESCO-main
delays.png 431KB
setup.py 1KB
maps.png 927KB
resco_benchmark
__init__.py 2KB
traffic_signal.py 12KB
states.py 11KB
main.py 5KB
utils
avg_duration.py 841KB
readXML.py 4KB
avg_waitingTime.py 843KB
avg_queue.py 848KB
graph.py 6KB
readCSV.py 2KB
avg_timeLoss.py 863KB
rewards.py 7KB
agents
mplight.py 5KB
__init__.py 210B
maxwave.py 2KB
pfrl_ppo.py 3KB
agent.py 3KB
maxpressure.py 738B
pfrl_dqn.py 7KB
stochastic.py 667B
ma2c.py 25KB
fma2c.py 7KB
environments
ingolstadt7
LICENSE 34KB
ingolstadt7.rou.xml 296KB
ingolstadt7.net.xml 267KB
ingolstadt7.sumocfg 213B
cologne1
cologne1.rou.xml 178KB
cologne1.sumocfg 207B
Creative Commons Legal Code.pdf 182KB
cologne1.net.xml 38KB
ingolstadt21
ingolstadt21.sumocfg 215B
LICENSE 34KB
ingolstadt21.rou.xml 415KB
ingolstadt21.net.xml 1.77MB
LICENSE 85B
arterial4x4
arterial4x4.zip 23.73MB
Creative Commons — Attribution-NonCommercial-ShareAlike 4.0 International — CC BY-NC-SA 4.0.pdf 135KB
arterial4x4.sumocfg 213B
arterial4x4.net.xml 154KB
ingolstadt1
ingolstadt1.rou.xml 168KB
LICENSE 34KB
ingolstadt1.net.xml 32KB
ingolstadt1.sumocfg 213B
cologne8
cologne8.rou.xml 184KB
cologne8.net.xml 346KB
cologne8.sumocfg 207B
Creative Commons Legal Code.pdf 182KB
cologne3
cologne3.rou.xml 746KB
cologne3.sumocfg 207B
cologne3.net.xml 146KB
Creative Commons Legal Code.pdf 182KB
grid4x4
grid4x4.zip 18.85MB
grid4x4.sumocfg 205B
grid4x4.net.xml 451KB
Creative Commons — Attribution-NonCommercial-ShareAlike 4.0 International — CC BY-NC-SA 4.0.pdf 135KB
multi_signal.py 9KB
config
__init__.py 211B
agent_config.py 5KB
mdp_config.py 15KB
Environment Configuration.md 6KB
signal_config.py 47KB
map_config.py 4KB
Creative Commons — Attribution-NonCommercial-ShareAlike 4.0 International — CC BY-NC-SA 4.0.pdf 135KB
MANIFEST.in 50B
.gitignore 2KB
README.md 4KB
共 69 条
- 1
资源评论
快撑死的鱼
- 粉丝: 1w+
- 资源: 9154
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功