edit ： Chen Ping 、 Devil
SDK It can also play games automatically ？ This SDK somewhat 「 cool 」.
In recent days, , Tencent Entertainment （IEG） Open source a project called GAME AI SDK Automated test platform for , The platform encapsulates a variety of tools for developers to use , Currently, the types of games supported are Parkour 、 Eat chicken 、 Shooting category 、MOBA Class etc. .
Project address ：https://github.com/Tencent/GameAISDK
Platform built in 「 Cool run every day 」 Example . The picture on the left shows the untrained random action , The picture on the right shows the effect of training .
SDK（ Software development kit ） It's usually a software engineer for a specific package 、 software framework 、 Hardware platform 、 Operating system and other development tools that can be used to create application software .
It seems a little abstract . In the actual project , We just need to remember ,SDK It's provided by mobile game channel , Integrated user login 、 Community function 、 Social sharing 、 Function module of data background statistics function . Access SDK after , Game manufacturers and channels should be right SDK Package to test , Only after passing the test can we go online .
Look at the cool running picture shown above , Do you want to try chicken 、 Shooting game experience ？ This open source project can meet your needs , It supports users to access projects and secondary development .
AI SDK platform
AI SDK The platform is a platform for developing games based on game images AI Open Source Toolkit for . It's mainly done UI testing 、 In game element recognition 、AI Algorithm （DQN、IM） And so on . Developers can complete the game automation test based on this toolkit .
At present, the game types supported by the platform are Parkour 、 Eat chicken 、 Shooting category 、MOBA Class etc. . It can replace the artificial automation of the game scene in specific scenarios , Reduce the manpower investment in game scenario testing .
The toolkit mainly includes AI SDK、AI Client、SDKTool（ Profile tool ） These parts .
SDKTool Help us generate game specific AI SDK Required profile . For example, in the course of the game , Action interaction is needed in the whole process UI、 Game elements that need to be identified, etc .AI The information used in the development process can come from game images （AI Client obtain ） Or other data .AI SDK Able to automate the game UI Operate and enter the game scene , Operate the mobile phone to play games like a player .
GAME AI SDK The platform has the following functions ：
Provides a variety of image recognition algorithms , Identify key feature data in the game scene , Such as the skill state in the game scene 、 Button position 、 Target location, etc .
Provide imitation learning and reinforcement learning AI Algorithm , It can be directly used in training game scenes AI.
It provides a complete set of sample collection 、 Image annotation tool .
AI SDK Module introduction
AI SDK It mainly includes AI Client modular 、MC modular 、IO modular 、UI modular 、GameReg modular 、AI modular .
chart 1：AI SDK Module diagram .
AI Client It mainly interacts with the mobile terminal , Get game images , And implement UI or AI Output action ;
IO Module and AI Client Interaction , As AI SDK Data input and output control ;
MC Module and IO Module interaction , Distribute data to AI SDK Back end UI and AI modular , And do some business logic management control ;
UI Module based on game image , Identify what needs to be done UI operation ;
GameReg Module based on game image , Identify the numbers in the game scene 、 Blood strip 、 Objects and other elements ;
AI Module based on the identified game elements , according to AI Algorithms perform business logic .
chart 2： Architecture diagram .
The deployment environment
function AI SDK Medium demo, First you need to install the dependency , And then local PC Mobile phone connected to mobile phone （ Or mobile phone emulator ）, adopt AI Client Real time collection of mobile phones （ Or mobile phone emulator ） Game screen of , And to the AI Service processing . hold AI The action result of service processing is fed back to the connection PC On my cell phone , Complete the interaction of actions .
chart 3：AI SDK Interaction diagram with mobile phone .
Deployment depends on the environment
The deployment dependent environment is mainly installation AI SDK、SDKTool（ Configuration file generation tool ）、AI Client The environment depends on .
The project provides two kinds of operating system deployment process .
Ubuntu： at present SDK The toolkit runs on 16.04 Of 64 position Ubuntu On the system , The deep learning framework used is TensorFlow, So before you use it, you need to install the dependent ubuntu The system and the corresponding dependent third party toolkits , Such as TensorFlow、OpenCV、protobuf etc. .
Besides , The project also introduces the profile tool SDK Tool、 Project access 、 Secondary development and debugging skills , Please refer to the project link for details .
Reference link ：