The condition that the vehicle is more than 3m away from the circle line
was too sloppy. That often happens when the radius is changed by sticks.
A reapproach is only necessary when the center is moved and that's only
possible through the orbit command.
The initial approach to the circle to orbit on was very agressive since
it was just the controller trying to stay on the circle reaching the
limits. Now there's first an approach phase in which the vehicle reaches
the circle trajeectory in a smooth perpendicular line before starting the
orbit execution.
Before it was:
- not used anywhere
- copied from an old mission implementation version
- didn't plan in advance
- had a lot of broken cases
- dependent on a lot of parameters
I'm starting with a new relatively simple implementation that works as
expected for a minimum viable implementation and can be improved over time.
The first version is used to approach the circle path in Orbit mode to
verify the interface and get testing such that it gets eventually used
everywhere.
It was broken because _velocity_setpoint is used for input and output
and was assumed to be the input at a place where it was already overwitten
To clarify this, the input setpoint is renamed "target"
Also remove crosstrack P controller that produces overshoots when the
acceptance radius is large (crosstrack error is suddenly large at
waypoint switch).
This is done to allow proper initialization of the new FlightTask and
give it a chance to continue the setpoints without discontinuity. The
function checkSetpoints replaces the setpoints containing NANs with an
estimate of the state. The estimate is usually the current estimate of
the EKF or zero.
The transition FlightTask also provides an estimate of the current
acceleration to properly initialize the next FlightTask after
back-transition. This avoid having to initialize the accelerations to
zero knowing that the actual acceleration is usually far from zero.
the vehicle yaws towards the next waypoint before accelerating. This is
required for drones with front vision and aerodynamic multicopters such
as standard vtol planes or highspeed multirotors.
The linear mapping from position error to cruise velocity is changed by
a combination of that linear mapping and a nonlinear function containing
the maximum acceleration and jerk to avoid overshoots at waypoints due to
overoptimistic breaking distance.
This is done to avoid generating large yaw changes when the velocity
vector is small; for example when switching into loiter or reaching the
last waypoint.
Matthias Grob