WHERE THIS TEST ENDED
An early manual sequence reaches the dashboard, and I report a successful final voice attempt without filming it again. Boot timing and reliable command delivery still need work, and screen-reading navigation stays on the backlog.
- Voice request to the Echo
- Fletcher to the controller
- Infrared HDMI sequence
- Projected dashboard
How it works
The M5Stack Echo takes the voice request. Fletcher tells the Seeed controller to send infrared commands to the projector: power first, then the navigation keys needed to reach HDMI. The controller uses the codes captured from the projector's own remote.
I also wanted the AI HAT to read the projected home screen and choose the next remote command. The build records show that a local vision-language model could describe the screen, but it was poor at locating the highlighted tile. In the video I put that approach on the backlog and use the known key sequence instead. This was testing an existing model, not training a new one.
What remained open
Boot timing and reliable command delivery still mattered. The clip supports an early working attempt, not a claim that every start succeeds. It also distinguishes the on-camera sequence from the successful final attempt I report at the end.
Later follow-up
The September 7 release records add power-state checks using the HDMI connection, a wait for boot, and bounded navigation retries. They also lengthen the power-off confirmation window to avoid sending a second toggle while shutdown is still underway. Those safeguards describe the subsequent implementation, not proof that every trial in this video already used them. Navigation by sight remained on the backlog.
About this breakdown
Matched to this original TikTok, including its spoken captions. The result above describes this episode; later work is identified separately.
Breakdown checked September 6, 2026.