Autosport Labs

I noticed that in my saved sessions from Horse Thief Mile this past weekend, the logitudinal acceleration values were very noisy and unstable. I am thinking this may be due to the large angle of climb and descent during most of the track. It appears as if the long accel vector is not being normalized against the pitch of the unit, which causes the forces on the Z axis from track bumps to be interpreted partially as longitudinal acceleration changes. I attached a link to a Racerender overlay video showing what I mean. The Lat accel values are smooth as always. My Racecapture/Track MK2 was calibrated prior to the track day in my garage on a level surface. I also added a gauge showing the % of climb/descent. The steeper the grade gets, the noisier the long_accel data becomes, which you can see in the G-Force graph in the bottom right of the screen.

Any ideas?



https://youtu.be/kl30T5EiLDg

Hi, How noisy is the IMU when the unit is at rest? We want to rule out if there's anything wrong with the sensor, or, if you're simply actually getting that reading in the unit.

Are you seeing this at other tracks where things are a bit more level?

The IMU is stable when viewing the 3d accelerometer visualization page in the dashboard. I have attached a screenshot of the data. I also noticed that the Pitch change in degrees/sec is way out of whack. Also, there is no way my 140 whp Miata could accelerate at 1.4 g! Although it would be fun!

I'm still seeing this noisy Lon Accel data. I have also seen it on tracks that don't have steep climbs or descents. I have recalibrated the accelerometer multiple times with no change. The longitudinal G data is pretty much useless.

https://youtu.be/lq_rrMppsy4

Hi, thanks for the update, and sorry for the late reply.

The accelerometers and gyros only receive a light treatment of real-time filtering in the form of an exponential moving average filter, which may still let some high frequency noise through, which could make it look like you're seeing higher g-forces than realistic.

We're looking at some post-processing filtering, using a Savitzky Golay filter which will help retain the sensor's curve while filtering out noise.

We'll also look at exposing the tuning knobs on the real-time filters to see if that can help as well. With filtering, there's always the risk of introducing lag in the data, so fine tuning will be required.

Thanks!