资源算法deepdrive-universe

deepdrive-universe

2020-01-09 | |  33 |   0 |   0

DeepDrive in Universe

Pretrained self-driving car models that run in GTAV / Universe

Prerequisites

Follow the env setup instructions here

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 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, hitvuntil you see something like thisdeepdrive load

If you see the steering wheel, change the camera settings like so:deepdrive load

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

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 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


上一篇:Berkeley-DeepDrive

下一篇:DeepDriver

用户评价
全部评价

热门资源

  • seetafaceJNI

    项目介绍 基于中科院seetaface2进行封装的JAVA...

  • spark-corenlp

    This package wraps Stanford CoreNLP annotators ...

  • Keras-ResNeXt

    Keras ResNeXt Implementation of ResNeXt models...

  • capsnet-with-caps...

    CapsNet with capsule-wise convolution Project ...

  • shih-styletransfer

    shih-styletransfer Code from Style Transfer ...