LatestMUV-LUV GIRLS GARDEN Mellow Days Launches on Steam With 10% Introductory Discount
Gaming

miHoYo Develops Gamer AI That Can Play Honkai: Star Rail by Watching the Screen

miHoYo, the developer behind Genshin Impact, is developing a new artificial intelligence system called Gamer, designed to play video games by observing what appears on the screen. The technology has already been tested in Honkai: Star Rail, where it demonstrated the ability to engage in combat, explore environments, complete quests, and defeat bosses.

Unlike AI systems that interact directly with a game’s internal systems, Gamer operates through visual input. It analyzes screenshots to understand the current situation before deciding which keyboard or mouse inputs to perform. This allows the AI to control characters and interact with the game in a manner similar to a human player.

How miHoYo’s Gamer AI Works

To improve its understanding of gameplay, miHoYo’s researchers trained Gamer using gameplay recordings accompanied by instructions, explanations of important decisions, and memory summaries. With this information, the AI learned to perform various actions, including moving characters, adjusting the camera, opening menus, changing areas, and interacting with objects.

The researchers evaluated its capabilities through two Honkai: Star Rail side quests: Opportunity Makes the Thief (Part I) and Bestial Ferocity. Both quests required the AI to travel between locations, talk to characters, fight enemies, and solve simple puzzles.

Gamer achieved completion scores of approximately 80% and 83.6%, respectively, demonstrating that it can handle several gameplay objectives with relatively strong results.

Puzzle Performance Still Needs Improvement

However, Gamer’s performance was less consistent when dealing with puzzles. Across 40 tasks covering seven different minigame categories, the AI achieved an average success rate of just 49.7%.

Its performance varied significantly depending on the puzzle type. Gamer reached an impressive 95.8% success rate in Navigation Compass puzzles, but struggled with Abacus Circuitry, where its success rate dropped to only 22.2%.

These results suggest that while the AI can understand and perform many basic gameplay actions, more complex challenges remain difficult for it.

AI Could Help miHoYo Test Future Game Content

miHoYo does not view Gamer as a technology intended to replace human players. Instead, the company sees it as a potential development tool that could assist with quality assurance and automated testing.

Gamer could eventually help developers verify whether quests, interactions, and objectives work as intended. By automatically exploring areas and repeating specific actions, the AI could also help identify bugs more efficiently.

If further development improves its reliability, Gamer could reduce the time and cost required to test new content, giving miHoYo another way to streamline its game development process.

Discussion

Tinggalkan Balasan

Alamat email Anda tidak akan dipublikasikan. Ruas yang wajib ditandai *