01 / OBSERVE
Read the screen
The planned agent uses captured game video as its view of the world. The experiment is designed around visual feedback, without reading game memory.
Learning to act.
One experiment at a time.
Can an AI agent observe a game, try an action, and learn from what happens? We’re building a hands-on experiment to explore that question—and what it might teach us about robotics.
Starting with The Legend of Zelda: Breath of the Wild on a real Nintendo Switch.
Inside the experiment
Watch the research
Watch a recorded clip now, or switch to the live feed when a broadcast is available.
No live broadcast right now
The live feed is not connected yet. In the meantime, select Replay to watch the recorded capture check.
Recorded hardware setup check, not an autonomous research run. Silent clip. Use the player controls to play, pause, seek, or enter full screen.
AI game agent research
Observe. Act. Compare.
02 / ACT
Try a control
A Raspberry Pi Pico sends controller inputs to the Switch. The aim is to discover what actions do through trials and feedback.
03 / COMPARE
Learn from outcomes
We plan to record actions, observations, and results so that successes and failed attempts can inform the next experiment.
Project status · October 3, 2026
Building the foundations
Video capture and basic controller actions have been tested. The controller firmware has automatic input-release safeguards. Work is now focused on a reliable controller service and a first budget-limited AI run.
The autonomous learning loop is still in development. A recorded capture check is available in the player above. Live broadcasting and research-run replays are still being connected.
Future direction
From game worlds to physical systems
Observation, action selection, feedback, and recovery are also useful questions in robotics. We hope these experiments will help us investigate AI for mechanical systems. Transfer to real robots is a research direction, not a demonstrated result; physical motion will need its own testing and safeguards.
Get in touch
Contact me
Interested in AI agents, game experiments, or robotics? Ideas, questions, and potential collaborations are welcome.