BowlSense Learn

From the Workbench to Your Bowling Ball

A look inside several days of home calibration: what the sensor tests showed, what still needs work, and the planned app utility that will help keep sensors up to date.

If you’re waiting on a BowlSense sensor, you deserve to see what is happening behind the scenes. Over the last few days, a lot of that work has happened at a desk and on the floor at home: holding sensors still, turning a ball into known positions, rolling it forward and backward, and comparing the recordings with video.

We now have stronger evidence that the tested sensor can retain a recording through Bluetooth interruptions, a desktop correction that passed a fresh stationary check, and encouraging results after rotating the sensor inside the ball. We also found calibration assumptions and release detection mistakes that need work before these experiments become dependable app features.

This update covers the September 4–8 development work. It is a progress report for the hardware and software being tested, with the remaining work laid out below. It does not announce a shipping date. First, make sure the recording survives Before calibrating a measurement, we need to know the sensor actually recorded it correctly.

The latest desktop acceptance run used a Seeed XIAO nRF52840 Sense with its built in motion sensor and the experimental 2.1.0 duty cycle firmware. Earlier testing on the C6 development hardware helped explore mounting and sleep/wake behavior; the results below identify the Sense tests separately because results from one board do not automatically transfer to another.

During development we found an acquisition problem that appeared while Bluetooth was connected. The sensor could appear connected while its internal sample sequence went wrong. After changing how the firmware reads that data, the final desktop acceptance run finished with zero recorded FIFO errors—the errors that flag problems in the sensor’s sample queue.

We deliberately disconnected Bluetooth during recording, interrupted a download, reconnected, and downloaded again. The completed recording remained available, and repeated downloads matched byte for byte. The local workbench saved the recording before telling the sensor it could clear that capture. Resetting or losing power still clears this candidate’s RAM backed recording.

Battery powered still captures also succeeded, including one after 75 seconds disconnected and untouched. Physical motion wake was observed on USB power. Longer battery endurance, measured sleep current, and a separate physical motion wake check on battery remain open. A good reconnect test tells us something useful; it does not tell us how many league nights a charge will last.

What six still positions can teach us A stationary accelerometer should measure roughly one g from gravity, whatever direction it faces. Holding it in several directions lets us look for consistent offsets. We recorded six distinct, stable positions for 11 seconds each: 27,462 samples in total. Before correction, their average acceleration magnitudes ranged from 0.98502 g to 1.01743 g .

A recording that included repositioning was kept in the notebook but excluded from the calibration. The experimental correction estimates an offset and a common scale from the measured directions. Because these were hand placed poses, we did not pretend the opposing positions were perfectly aligned. This simple correction also does not solve every possible sensor error. Each point represents one stationary recording.

The first six recordings supplied the fit. The separate tilted recording did not. The vertical axis is deliberately zoomed around 1 g so the differences are visible; this is a stationary magnitude check, not a bowling accuracy chart. Then we froze the correction and collected a new tilted recording—a holdout , meaning a test that was not used to tune the answer. Its raw reading was 1.