Odom vs Map and Better Absolute Bearing Discussion 07/06/2022

[niwhsa9]

Odom vs Map:

• Use odom frame for gate, lets us avoid crazy filtering of gate points from percep since there is no GPS noise to deal with
• AR tags in the odom frame in general, lets us take GPS noise out of the relative localization estimate. AR tag localization problems are all relative, short-scope localization so it makes sense to do it in odom
• Use odom frame for spiral, plan the entire spiral in the odom frame. Try it out to make sure the drift over the duration of the spiral doesn't cause huge problems. If we can do spiral in the odom frame, the "arc" driving problem goes away because we don't rely on absolute bearing at all. It just becomes all about how well can we drive dead-reckoning
• Absolute bearing is largely unavoidable when working with GPS waypoints in the map frame, though its use can be minimized by smart use of the odom frame, possible switch to odom at short distance to minimize the effect of the arc affect with bad absolute bearing measurements
• Cross track error in the control loop always exists as a last resort

Better Absolute Bearing:

• Lets learn about madgwick and mahony filters so we can make educated adjustments as opposed to random trial and error

• Lets run the sensor fusion algorithm on our own processor. This might give us faster updates and more insight into whats going on. There are also potentially better sensor fusion algorithms out there

  • If we have our own filtering, we can run two filters. One that prioritizes globally accurate data (can be noisy), one that doesn't. Can increase importance of magnetometer
  • GPS bearing can get fused in too, the filter can be state aware too make sure GPS bearing only gets fused in when driving straight

• Lets dig into manual calibration profiles, such as motor throttle aware profiles

Misc:

• Add a wiki page explaining transform syntax
• Try out the ZED IMU solution, just in case its magic
• Obstacle detection comes free from surface normal based cost maps

This is a useful read: https://hackaday.com/wp-content/uploads/2019/03/hackaday_journal-gregorytomasch_kriswiner-heading_accuracy_using_mems_sensors.pdf