Use Sensor Fusion approach to detect stationary and filter out at motion model (i.e., needs to be moving to add to the model)

[Masaya-RT]

We want to filter out data where we have "dancing position" when car has GPS but is stationary

[Masaya-RT]

Kalman filter: Filter wasn't stopping the issue, but was reducing the error radius.

• Problem is and has existed, but filter helps

[roger-berman]

Two Problems:

1. Start of drive motion
2. During drive stopped motion

Need to operate within loop in motion model in odc api. Near real time processing to attempt to deal with motion vs no motion of vehicle.

Full solution should be implemented in odc api

• Test data can be grabbed from CTP group for any HDC data from the past 7 days, will have turned off kalman filter

[Masaya-RT]

Not many updates Question for Alexei: Start of the drives, you query sensor logs and see motion model. Is it filtered out there? Yes

Maybe EOD progress

[Masaya-RT]

Inital experimentation is good. Python is proven

Solid portion converted to TypeScript, but Python Math still needs migration. Trying to figure out low-pass filter implementation.

Needs to be widely-applicable and relatively noise-proof.

[Masaya-RT]

Everything at this point is written and tested in TypeScript. Performance is slower than Python Roger to discuss with Alexei to weave into motion controller (tradeoff implications) Should be ready for testing after

[Masaya-RT]

Gave Alexei PR. Waiting on first test results

• Will be packed for Masaya (just Masaya today)

[Masaya-RT]

Still waiting on feedback from the branch

[roger-berman]

Got Data from @punov and will be familiarizing myself with @mazar-farran's new test framework to begin debugging performance

[Masaya-RT]

Results still pending - Focusing on Data Logger refactor

[Masaya-RT]

Ran E2E test, was breaking.

• To talk to Mazar to figure out issue