The 2023 World Artificial Intelligence Conference (WAIC) will be held in Shanghai from July 6th to 8th. Leading scholars in the field of global artificial intelligence, multiple Turing Award and Nobel Prize winners, and representatives of well-known technology companies , representatives of international organizations, etc. gathered to share and discuss new achievements, new opportunities, and new challenges of artificial intelligence technology. On the third day of the conference, the "Game AI Application and Game Technology Forum" was successfully held under the guidance of the Shanghai Xuhui District People's Government, hosted by Tencent Games and Tencent East China Headquarters, and supported by Shanghai West Bund Development (Group) Co., Ltd. Many guests from the game industry attended the meeting to discuss the technological accumulation and vision of AI technology in the game industry, showcase the innovative expansion cases of AI technology in different game categories, and the application and development of AI+ game tools in cross-border fields.
In this forum, Yu Dong, CoDM of Tianmei J3 Studio of Tencent Games and head of the development of the mobile game Ni Zhan, shared the topic of "AI Application and Cross-Field Scientific Innovation Exploration in FPS Games". It focuses on the research and development breakthroughs of J3 Studio in anthropomorphic AI and intelligent agent action generation technology for FPS games, as well as the application and imagination of related technologies in cross-domain technology innovation.
(Yu Dong, CoDM of Tianmei J3 Studio of Tencent Games and head of mobile game development of Counter War, shared the scene)
Ten years of deep cultivation in the FPS field J3 layout New breakthroughs in category technology
Tianmei J3 Studio is a shooting game research and development studio under Tencent Games Tianmei Studio Group. It has brought players a variety of familiar gun battle games and accumulated solid R&D strength in the FPS field. . In 2015, J3 launched "CrossFire Mobile" and quickly captured the mobile shooting game market upon launch. In 2019, "Call of Duty Mobile" jointly developed by J3 and Activision Blizzard was released globally, and has successively won a number of domestic and overseas industry awards, including TGA's Best Mobile Game of the Year. As early as 2012, the self-developed shooting team of the predecessor of the J3 team, Linlang Tianshang Studio under Tencent Games (later merged with Tianmei Yiyou Studio), launched Tencent's first self-developed full-mode shooting online game "Reverse War". It is also the leading domestic self-developed FPS online game. Currently, there are still many new games in the development process of J3. In 2022, the new UE4 science fiction PVE shooting game "Counter War" IP of "Counter War" IP will also be exposed for the first time.
(Online game product of Tencent Tianmei J3 Studio)
(In-development sci-fi shooting PVE product "Reverse Fighter" "Game")
Speaking of AI technology, especially the development of deep learning, it has undoubtedly opened up many new possibilities for the game industry's research and development process and player experience improvement paths. However, because FPS games often use 3D stereoscopic scenes, and the game has the characteristics of complex environment, incomplete field of view, many enemies and friends, and high diversity of environmental utilization, the AI ??technology research in FPS is orders of magnitude more difficult than that of 2D and 2.5D games. increase. With these challenges in mind, J3 teamed up with Tencent's AI expert team to launch the first attempt of deep learning in FPS games.
Working with Tencent AI Lab to develop industry-leading FPS anthropomorphic AI
Since 2018, J3 Studio has joined hands with Tencent AI Lab to launch a humanoid AI Technology exploration with AI as the core goal. Faced with the challenge of AI learning complexity brought by the 3D environment, Tencent AI Lab and the joint team on the game side have gradually overcome technical difficulties such as 3D environment perception personification and 3D operation personification, taking the lead in achieving industry-leading capabilities that can solve long-term problems. Full-image FPS AI solution for sequence decision-making, long-distance planning, and cover transfer and utilization capabilities.
In 2020, "CrossFire Mobile"'s first new story interaction + competitive mode "E-Sports Legend" will be launched. The team uses AI to analyze a large number of real games played by professional players and simulate and restore their operations in the game. In this way, CF mobile game professional players Guobao, Manman, Akan and others have been integrated into the game one after another. Players can have a passionate and immersive 1v1 game with these masters at any time.
In 2022, "Call of Duty Mobile" will also launch a 5v5 e-sports challenge mode. The Q9 team that won the Professional Masters Championship that year was customized as an AI character in the game. Players can form a team to challenge the "AI clone" team trained based on professional player data, and experience the competitive fun of the peak showdown in the e-sports arena.
Cooperating with Tencent Robotics The PvE shooting product "Reverse War Mobile Game" under development has made certain research and development breakthroughs in the field of AI action generation technology. In the battle, growth and exploration in PvE games, the performances and battles of a large number of bosses, mobs, NPCs and other characters are crucial. These are core components of providing players with an immersive content experience. The traditional action solutions in the industry are mainly state machines, relying more on the stacking of huge behavior trees and animation resources; in recent years, overseas manufacturers have begun to implement more advanced and efficient generation solutions, such as Red Dead Redemption using action matching technology ( Motion Matching), HyperMotion used in FIFA 22 and other cutting-edge explorations.
The NiZhan mobile game team focuses its research on action generation on the ARNN (Auto Regressive Neural Network) model, which is an autoregressive neural network algorithm model. This is an algorithm widely used to analyze and process time series data, and is often used in scenarios such as weather and stock market predictions. Simply put, if it can be used in the action field, it can be trained with a large amount of real-person motion capture input data, so that the AI ??model can predict the next frame of the action sequence frame, and finally generate complete and natural actions in a coherent manner.
ARNN (Auto Regressive Neural Network) model
Starting from 2019, Counter War Mobile Games has joined forces with Tencent Robotics X Laboratory to conduct research based on the ARNN model. Started and gradually completed the research and development of "intelligent action generation technology", a cross-cutting cutting-edge technology in the fields of games and robots. This technology is based on the principle of the ARNN model, learns motion capture data, and uses data-driven adaptive algorithms to allow NPCs or robots to automatically generate actions, reactions and expressions that are more like real creatures based on different external reactions such as player operations or environmental changes. .
This is expected to be the first real application of the autoregressive neural network algorithm in the field of online game action generation. During the research and development process, the biggest difficulty was how to apply the algorithm to the online game production ecosystem: the team built a supporting tool chain from scratch and followed the three steps of motion capture, animation preprocessing, and continuous training and tuning. A lot of investment and research has been done. Faced with the problem of network synchronization and performance optimization of AI action generation in online games, developers have adopted trajectory matching technology, fusion animation and corresponding algorithm-level solutions to optimize the overall effect performance and neural network reasoning overhead. Currently, it has reached a very high level. Good performance test results.
For "Reverse War Mobile Game", the use of new technologies will be accompanied by more natural performance of basic in-game actions and a more exciting and immersive combat experience. At the same time, this technology can also help in dimensions such as optimizing player package size and improving R&D productivity.
At the scene, Yu Dong shared some technical effects. It can be seen that in the multi-mob chasing effect animation using this technology, the mob character movements generated by AI in real time are very vivid and diverse.
An illustration of the chasing effect of multiple mobs on the same screen using agent action generation technology
In terms of the cross-industry application of agent action generation technology in the field of robotics, There are also very good performances and progress. This technology is also used in the self-developed quadruped robot Max developed by Tencent Robotics X laboratory. On the one hand, it helps the robot improve its thinking ability, allowing it to achieve more natural and smooth motion trajectories, and intelligent generation, control and decision-making of finer-grained actions under continuous training; on the other hand, with the help of real-time physical simulation technology, the research team Realistic, complex and diverse robot virtual training scenes can be quickly established in the game. Coupled with the acceleration capabilities of the game engine, the robot training process and time can be greatly shortened.
Tencent independently develops the quadruped robot Max
Currently, there are several patent applications for intelligent body action generation technology in the process. In the future, in the field of games, this technology may be reused to generate overall actions in more realistic-style games, promoting the dual progress of industry technology and user experience; and in the real world outside games, with the further progress of robot self-research, As it advances, it may also have the opportunity to create greater value in cross-field industries such as scientific research, elderly care, and citizen services.
Use game technology to make robots more "smart"
Imagine the new FPS AI ecosystem and look forward to the new era of artificial intelligence
In this forum, Yu Dong also expressed some thoughts on the future development of game AI technology.
One of the ideas is AI automated testing in the research and development process. The degree of test automation is a very important indicator in the DevOps field, but it is more difficult to implement industrial software-level automated testing for PvE games with complex logic and rich content.
But in the future, when AI gradually has the ability to perceive the environment and atmosphere, real-time analysis capabilities, and the ability to make independent decisions about test boundaries and conditions, perhaps this scenario will come true: input a task, and AI can Execute instructions and explore the experience in the game like a real person, and generate detailed test and feedback reports. This will greatly improve the quality and efficiency of game testing and help ensure and improve game quality.
Another fantasy is about the AI ??NPC ecology, which is also a very hot topic in the industry recently. For example, this year there was a paper titled Generative Agents, which described a "virtual agent" project by researchers from Stanford University and Google. The experiment of "autonomous survival" of 25 AI agents in "Town" has aroused great attention and discussion in the industry. When AI gradually develops and can possess more enhanced natural language, behavioral decision-making and social capabilities, perhaps building a true AI NPC self-running ecosystem in the virtual world will no longer be a fantasy.
Generative AgentsGenerative Agents: Interactive Simulacra of Human Behavior
The new generation of games is already at the forefront of change. AI has brought a vast space of imagination to the development and creation of games. We also look forward to the evolution of technology, domestic games can continue to promote cutting-edge exploration and achieve breakthroughs, and create more unique, real and deep game virtualization. world. And by that day, we also believe that game AI+ technology will definitely break the traditional scenario and bring about the possibility of unlimited cross-field technological innovation in the real world.