AI technology helps astronomical technology, optimize the satellite constellation intelligent observation system

On June 27, 2022, Tencent Games announced that it will use artificial intelligence (AI) technology to help the Chinese Academy of Sciences High -Energy Penal Physics Key Laboratory "Catch: Chasing All Transitions Constellation Hunters" program. Push my country's space astronomical observation technology to a new height, and make important contributions with the observations of time domain astronomy.

The Catch plan is the intelligent X -ray astronomical constellation of the key laboratory of particle celestial physical fitness and consisting of hundreds of microfinee. It is planned to be fully deployed around 2030. Its core scientific goal is to "portray the multi -dimensional dynamic panoramic view of the extreme universe" Essence The cooperation between Tencent Games and particle celestial physics key laboratories is committed to promoting the application of multi -intelligent learning algorithms in the collaborative observation of the Catch constellation. At that time, Catch plans to use Tencent Games to lead the industry's leading AI technology to realize the functions of constellation independent collaborative observations.

Tencent Game CROS AI R & D team trained in the game has reached the level of top players; the team also innovatively launched applications such as human -machine collaboration and human -machine confrontation. Essence With the continuous development of AI technology, human-machine collaboration has achieved good application effects in the past exploration, and the "understanding-communication-collaboration" method also helps the system to achieve the established goals more efficiently.

CatCh constellation of the imagination of the target celestial body under AI technology cooperates

Just as the AI ​​intelligence developed by the Tencent Game CROS AI R & D team, its intelligence and agility originated from the experience of studying and precipitating from hundreds of millions of "human -machine model" game battle; Learn "past astronomical observation data. At present, the algorithm engineers of Tencent Games combined with the explosive source data of Chinese astronomical satellites, they initially set up explosive source simulation simulators to train AI algorithms exclusive to the space observation environment.当CATCH星座在太空中运行时，将借助上述算法对深空中成千上万的爆发源数据进行实时分析，调度卫星执行目标选择、指向调整、编队组合等观测指令，对观测目标进行全天、 Full -time monitoring will be optimized for continuous algorithm based on the results of observation.

How to control hundreds of satellites to observe the explosive sources and transformer celestial bodies in the universe more efficiently? As early as the concept of Catch, scientists realized such a challenge. From the traditional experience, the astronomical observation of a single satellite often requires a special team to run; for the CATCH constellation of up to 100 satellites, this seems to be an impossible task. To this end, the Tencent Game CROS AI R & D team has fully discussed with experts in space and astronomical aspects, and put forward more ideal solutions: using the latest training technology of the game AI -multi -smart body strengthening learning methods to control satellites to control satellites Cooperate and cooperate to complete various observation tasks and optimize observation strategies.

Specifically, the team is equipped with a high -combined distributed environment for model training, which can support the access to the tens of thousands of simulation environment training. It can also use training samples for distributed training. At the same time Deploy the optimization plan in order to effectively balance the perception of space signals, the requirements of control accuracy, and the restrictions of satellite computing power, and efficiently complete the collaborative observation tasks of multiple satellites. This plan not only includes the accuracy requirements of scientific research purposes, but also in line with the computing power requirements of real -time dispatching of the space environment, but also support the basic system architecture requirements of large -scale computing power.

Tencent Game AI Multi -Smart Body Algorithm has been applied to multiplayer online tactical competitive games such as League of Legends mobile games and Naruto mobile games. Specifically, the game AI can match the level of deep learning and strengthening learning training through multi -intelligence, so that its own battle level can match the level of most players. Under the "decentralized" multi -smart body algorithm scheduling, the game AI can not only cooperate with the player, but also enable players to experience different styles of tactics. In addition, in training for game AI, improving the output intensity is not a single training goal, but a series of quantitative indicators are needed: from combat capabilities, defensive ability, to the degree of cooperation with teammates, etc. Gaming strategy and continuously optimize training parameters to improve the comprehensive performance of game AI.

"Tencent's R & D personnel quickly understood our research focus and showed a high interest in our research direction." Catch planned person and young scientist Tao Lian's cooperation with Tencent games, "If you can refine The valuable part of the game technology is correctly applied to all walks of life. This is a good thing for the development of the country and the development of technology. "

At present, the force of game technology is "spilling" to the real world, and the shape and boundary of the game are constantly upgrading and breakthrough. One year after the "Super Digital Scene" cognitive upgrade, Tencent Games also consciously applied game technology to cross -border game technology in "industry", "culture", "scientific research" and other reality areas, and continued to explore the diverse value of game technology.

The AI R & D team of Tencent Game CROS R & D and Effective Department, the business scope covers game AI R & D and intelligent NPC production, art automation, physical simulation, action capture and other frontier technology fields.The team has supported a variety of game products such as League of Legends, Naruto Mobile Games, QQ Speed, etc.Among them, intelligent robots and smart NPCs (Non-Player Character non-player characters) based on multi-intelligent deep enhancement technology production have formed a mature AI production platform, which can greatly improve research and development efficiency and optimize user experience.The team continued to study on cutting -edge academic research, and related research results papers have been published at academic conferences such as Neurips, CVPR, AAAI, Siggraph.Edit: Photon

- END -