Chinese studio miHoYo (known for the hit Genshin Impact) has introduced an artificial intelligence model called Gamer, capable of playing video games independently. The developers taught the algorithm to fight enemies, solve puzzles, and complete quests in another popular game from the company — Honkai: Star Rail.

During the video demonstration, the AI freely explored game locations, engaged in battles with opponents, and even managed to defeat a boss, reports Dexerto.

How the technology works

Unlike traditional game bots that interact directly with the game's internal code, Gamer behaves like a regular player.

The system takes screenshots of the screen to understand what is happening in the game, after which it issues standard mouse and keyboard commands.

Researchers combined gameplay recordings with text instructions, explanations of key decisions, and brief memory summaries. This allows the AI to keep current goals in mind, freely control the character, rotate the camera, navigate menus, and interact with the environment.

First test results

To test Gamer's skills, the developers ran it through two side quests in Honkai: Star Rail — Opportunity Makes the Thief (Part I) and Bestial Ferocity. They require movement between locations, dialogues with characters, battles, and solving basic tasks. The AI successfully completed 80% and 83.6% of the objectives in these quests, respectively.

At the same time, the model did not handle the puzzles as evenly — the average success rate was 49.7% on 40 tasks across seven types of mini-games:

  • in the Navigation Compass puzzles, the AI achieved 95.8% success;
  • in more complex Abacus Circuitry tasks, the result dropped to 22.2%.
  • Why the studio needs this

    At miHoYo, they note that the technology is being created not to replace gamers, but to automate game testing. The AI's ability to independently explore locations and complete missions will help developers detect bugs, verify the correct functioning of quests, and reproduce errors before updates reach the main servers. This will allow for a significant reduction in development and quality assurance (QA) costs.