Talk to Sid about existing EKF

[smriti-kumar]

Meeting Notes

• EKF right now is 2d, biggest goal is to transition it into 3d
• Explore error state kalman filters, not much clear information online
• Look into unscented kalman filters (don't need to take jacobian), ideal for parallelization (look into this)
• Start with a basic cartesian kalman filter simulation
• Read about the bicycle model in order to understand ackermann model, makes assumption that there's no axle between the front wheels
• Make a sim with ackermann model and ekf
• Do dead reckoning with wheel encoders and steering on mini car to make sure ackermann model makes sense
• Incorporate ramps with IMU pitch with above step
• Confirm what sensors are available to get data from
• Transition into testing the kalman filter I write, will need to tune covariances for the different sensors through trial and error (ex. IMU might be really bad at velocity)
• Make sure to set a ground truth when finetuning, ex. tape the course the car will take

[lucaslibshutz]

@smriti-kumar this is great! let's catch up on saturday and see if we can make a plan on how to move on this :)