English: The MAVeN navigation algorithm used has no absolute references to any landmarks. It always operates against a base frame where it sees a bunch of features and tracks them over a limited set of search frames. When it’s done, it requires a completely new base frame. It is always tracking in a relative sense, never tied back to a global frame.
MAVeN is implemented as an Extended Kalman Filter (EKF) that also uses the difference between the predicted and measured LRF range.
MAVeN has a state vector with seven components: position, velocity, attitude, IMU accelerometer bias, IMU gyro bias, base image position and base image attitude, for a total of 21 scalar components.
MAVeN only tracks features between the current search image and the base image. Because the base frame is frequently reset as the features are lost, MAVeN is effectively a long-line visual odometry algorithm: the relative position and attitude between the two images are measured, but not the absolute position and attitude…
The two main disadvantages of MAVeN are sensitivity to rough terrain, due to the ground-plane assumption, and long-term drift in position and heading. For Ingenuity technology demonstration phase, this is an acceptable tradeoff, because accuracy degradation is graceful and the algorithm has proven to be highly robust in both simulations and experiments.