# DeepDrive in [Universe](https://universe.openai.com/)
_Pretrained self-driving car models that run in GTAV / Universe_
## Prerequisites
Follow the env setup instructions [here](https://github.com/openai/universe-windows-envs/blob/master/vnc-gtav/README.md)
## Setup
```
git clone https://github.com/deepdrive/deepdrive-universe
```
Note: You will need GPU acceleration to do inference at the 8Hz that the model runs. Slower (or faster) inference may work, but is not the standard way the model is run.
### Baseline model - Caffe version
Install the latest version of Caffe and download the model via
```
cd drivers/deepdrive
wget -O caffe_deep_drive_train_iter_35352.caffemodel https://goo.gl/sVAedm
```
### (Work in progress) Baseline model - Tensorflow version
_Thanks to [Rafal jozefowicz](https://github.com/rafaljozefowicz) for helping train this model_
```
cd drivers/deepdrive_tf
wget -O model.ckpt-20048 https://goo.gl/zanx88
wget -O model.ckpt-20048.meta https://goo.gl/LNqHoj
```
Baseline models were trained with the standard hood camera in GTAV.
To enable the hood camera, hit <kbd>v</kbd> until you see something like this
![deepdrive load](https://www.dropbox.com/s/q28tce40ukurm9p/Screenshot%202016-10-30%2014.33.50.png?dl=1)
If you see the steering wheel, change the camera settings like so:
![deepdrive load](https://www.dropbox.com/s/h3xu98jz45bafld/Screenshot%202016-10-30%2014.28.42.png?dl=1)
In addition, set the resolution to 800x600 to minimize network I/O and enable borderless mode to avoid sending the window chrome through the network
![borderless](https://www.dropbox.com/s/dci8o6z3129bwpl/borderless.jpg?dl=1)
## Run the model
```
python main.py -d [DeepDriver|DeepDriverTF] -r vnc://<your-env-ip>:5900+15900
```
## Sample data returned by the env
```
{
"body" :
{
"done" : false,
"info" :
{
"center_of_lane_reward" : 0,
"distance_from_destination" : 1306.6157153344816,
"forward_vector_x" : -0.9870644211769104,
"forward_vector_y" : 0.15973846614360809,
"forward_vector_z" : 0.013689413666725159,
"game_time.day_of_month" : 6,
"game_time.hour" : 16,
"game_time.minute" : 8,
"game_time.month" : 9,
"game_time.ms_per_game_min" : 2000,
"game_time.second" : 47,
"game_time.year" : 2009,
"heading" : 80.808067321777344,
"is_game_driving" : false,
"last_collision_time" : 1481999559,
"last_material_collided_with" : "4201905313",
"on_road" : true,
"script_hook_loadtime" : 1481939095,
"speed" : 0,
"spin" : 0,
"x_coord" : -2372.70068359375,
"y_coord" : 1032.6005859375,
"z_coord" : 195.53288269042969
},
"reward" : 0
},
"headers" :
{
"episode_id" : "1",
"sent_at" : 1481999938.4742091
},
"method" : "v0.env.reward"
}
```
## Directly connected machines
To connect your Windows and Ubuntu machines directly via ethernet, follow [these instructions](http://askubuntu.com/a/26770/158805) for adding the interface to Ubuntu.
Use the _Netmask_ provided by `ipconfig` in Windows for your ethernet interface.
## Contributors
Several people, without which, this project would not have been possible
```
Aitor Ruano
Alec Radford
Alex Ray
Andrej Karpathy
Andrew Gray
Brad Templeton
Catherine Olsson
Christina Araiza
Dan Van Boxel
Dario Amodei
Felipe Code
Greg Brockman
Gustavo Youngberg
Ilya Sutskever
Jack Clark
James Denney
Jeremy Schlatter
Jon Gautier
Jonas Schneider
Krystian Gebis
Lance Martin
Ludwig Pettersson
Matthew Kleinsmith
Matthew O'Kelly
Mike Roberts
Paul Baumgart
Pieter Abbeel
Rafał Józefowicz
Tom Brown
Vicki Cheung
```
好家伙VCC
- 粉丝: 2500
- 资源: 9138
最新资源
- 西门子水处理程序学习手册:涵盖PLC程序、通讯点表、CAD原理图、操作说明及触摸屏功能介绍,西门子水处理程序 包含1200PLC程序,通讯点表,CAD原理图,操作说明 是学习污水处理的最佳案例 触
- 2cd36c672d44edff9777d9256b1ca68c.part28
- 基于MATLAB的单相PWM可控整流器仿真模型:快速响应、精准控制,实现网测电压电流同相位与软启动功能,单相PWM可控整流器,matlab,仿真模型,响应速度快,控制精度高,网测电压电流同相位,功率因
- 三相逆变并网与单相逆变的双闭环与单闭环控制策略解析:dq坐标变换、解耦与SVPWM和SPWM调制技术,三相逆变并网 离网,dq坐标变,PI双闭环控制 单闭环控制,解耦,svpwm 单相有源逆变 单相无
- 2cd36c672d44edff9777d9256b1ca68c.part29
- 汇川H3U标准程序:三轴定位脉冲控制,总线伺服定位与模块化控制优秀案例,汇川H3U标准程序,程序有本体脉冲控制的三轴定位,有总线控制的汇川伺服定位,轴点动,回零,相对定位绝对定位,程序结构清晰,分模块
- Qt调用FFmpeg库实时播放UDP组播视频流
- 三菱FX3U六轴标准程序:实现3轴本体控制与3个1PG定位模块联动,轴点动控制、回零控制及定位功能,结合气缸与DD马达转盘多工位流水作业模式,三菱FX3U六轴标准程序,程序包含本体3轴控制,扩展3个1
- 三菱FX3U三轴伺服电机与威纶通触摸屏集成程序:核心功能解析与编写入门,三菱FX3U三轴伺服电机程序,威纶通触摸屏程序,包含轴点动,回零,相对与绝对定 位,整个项目的模块都有:主控程序,复位程序,报
- kdeclassic-cursor-theme-4.11.19-16.el7-9.x64-86.rpm.tar.gz
- 基于FPGA的多通信接口设计:以太网、UDP IP、千兆网络、Uart串口与USB通信实现及资料支持,基于FPGA的以太网、UDP IP、千兆网络、Uart串口、USB通信接口设计 本设计是在FPGA
- kdegraphics-strigi-analyzer-4.10.5-3.el7.x64-86.rpm.tar.gz
- 基于FPGA的DDS原理信号发生器设计:利用Quartus II 9.1与Verilog HDL实现频率幅度可调的正弦波、方波、锯齿波及三角波生成器,包含代码与原理图 ,基于FPGA的DDS原理信号发
- 2cd36c672d44edff9777d9256b1ca68c.part30
- kdegraphics-thumbnailers-4.10.5-3.el7.x64-86.rpm.tar.gz
- 易语言AI文字识别模块:本地免字库高效识别,断网多线程轻松应对,源码示例无接口上传便捷操作,易语言AI文字识别,免字库本地识别无限制使用,断网可用,非搭建服务器那种,可以多线程使用,抛去了调用接口上传
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈