Easter Weekend. Amanda’s birthday get-together, ComedySportz, and Family. It’s finally warm outside on a consistent basis. It almost feels like the weather will build slowly to seek revenge for my winter bitching, with a finale of dry abundant heat, brought on by the Utah desert in August.
We’ve clocked in nearly 2,000 miles in on Trinity, with a few hundred miles of those being self-driving. Total fuel cost is $37.00 with an average of 73 MPG once the initial 30 electric-only miles are consumed.
I’ve been working with the Openpilot Community to sort out lateral control for the Prius. I fell as if I’ve crashed coursed into PID control engineering, learning the Python codebase as I went. I was considering going into detail into some technical specifics, but I’ll leave you with these two graphs.
The important metric is Actual Angle, represented by the yellow line. This is what the steering wheel is doing. The top is “expected” OP behavior, the bottom is my tuning. The zero boundaries are steering wheel center. Both plots were captured in a straight line.
So what if Kd overshoots validation logic? The end result is a smooth angle output. I’ll be entertaining a proposed reverse logic, a lower steerRateCost with lower Kd, Ki; but if it fails, you can bet I’ll be sticking with my lowered noise floor even if it’s high-frequency invalidation.