ENGINEERING CASE / ULTRASENSE EXPERIENCE
Two sensing problems. Separate setups. A development loop that turns measured behavior into the next firmware or algorithm candidate.
Hao-Yen Tang’s engineering work at UltraSense · October 14–21, 2026
A sensing algorithm needs repeated contact with the physical world. The useful change is to connect AI to the code, instruments and test procedures—then let separate setups carry separate lines of investigation forward.
Setup A / Force sensing
Firmware iteration and repeatable press campaigns
Results guide the next candidate
Setup B / Ultrasonic sensing
Filter research followed by firmware integration
Results guide the next candidate
Concept workflow based on the two development tracks. Engineers set objectives, prepare hardware and choose tradeoffs.
Each physical setup is an independent development resource. Sessions sharing one rig take turns controlling it.
Automated campaigns continue through configured test conditions between engineering decisions, so the next workday starts with fresh measurements and comparisons.
An automated press platform and a multi-axis test platform connect fixtures, sensors and control electronics to repeatable physical experiments.
Choose the behavior and the measurement that will show improvement.
AI works with the existing code, knowledge and recorded data.
Run software checks, load the firmware and execute the supported test.
Compare the measurements, choose the tradeoff and refine the next candidate.
The force-sensing track connected firmware development to an automated press rig. Seven firmware images were flashed on October 20. Separate in-air soak tests tracked stability as the algorithm evolved.
The reported 99th percentile of absolute force output fell from 5.829 N to 0.054 N across the shown development revisions. These separate in-air soak tests used different durations and tracked output stability while the probe was unloaded.
In the selected 10-round campaign, all 200 presses at or above 0.5 N were detected, with no false triggers reported in that campaign. It comprised 90 press runs and 10 soak tests. A later, broader validation found a missed first press and an early release during a light long hold.
The four recorded examples show fast taps, a slow ramp, light presses and a 30-second hold. The light-press example sits near the detection boundary. During the long hold, the output decays and the button releases at the configured 15-second limit.
The ultrasonic track compared 27 filters on recorded chip data, considering signal quality and estimated processing cost. A filter feature was merged into firmware the day after the study.
Recorded-data comparison; CPU time is estimated under two arithmetic-cost assumptions. The integrated feature was measured separately at an additional 31 µs per channel in firmware.
Start with one repeatable engineering task. Connect the knowledge, code and equipment, agree the checks and leave the team with a workflow it can use. Extend to another independent setup when the next experiment calls for it.
Development snapshot · October 14–21, 2026 · Hao-Yen Tang’s engineering work at UltraSense.