No autopilot will out perform a human helmsman.
No human helmsman can run an autopilot into the ground.
A human helmsman's brain will turn to mush before the autopilot's will.
Some helmsmen have turned to mush and don't know it.
The autopilot having turned to mush, will.

Like the Tin Man, the Scare Crow and the Lion, an autopilot needs 3 things.
A brain so that it can prestidigitate.
A strong arm to hang ten.
A star to steer by.

In practical terms this means the motor that does the work has to be
as strong as you or me. In horsepower terms consider how many foot
pounds of torque you can muster and for how long. If you undersize
the motor the system is crippled at the very beginning. 1/4 hp is
about the average of what most of us could muster over a long period.
Figure in the amount of electricity that it will take to feed the
motor and make sure it is continuous duty rated. More power will take
more electricity, at least when the motor is under increased loads,
but will be able to keep up better in extreme conditions.
If you can't steer the boat by hand under some arbitrary conditions,
in general the autopilot won't be able to either, due to hull, rudder
and other factors.
The autopilot can't steer to a heading unless that heading is truly
represented to it in real time. Any and every delay between the
actual heading and the heading that is in the brain box will result
in unresponsiveness, unrelated to the amount of power that is
available to turn the rudder. In other words, you can run a test by
attempting to steer by hand using the autopilot heading numbers and
looking at nothing else(get someone else to do the lookout). If you
can't steer to the heading, the autopilot most likely can't either.
In which case the heading indicator (compass) is most likely the
problem. It may be too small, it may have it's signal interfered
with. There are myriad possibilities here.
Just keep in mind that the autopilot brain is not getting it's
reference from sighting on a star.

The scare crow wanted a brain and so does the autopilot. But, there
are better brains and worse ones and there are myriad ways to
programming the autopilot brain. The best ones can be tweaked while
they are running without disabling the auto function. With this type,
feedback is immediate and you can generally manipulate the gain and
counter rudder quickly to suit any condition that you can encounter.
Assuming that the motor is strong enough and the heading update is
quick enough and accurate. There are far more of the worse kind in
service than the better kind.

Mike

Capt. Mike Maurice
Tualatin(Portland), Oregon

In my opinion, autopilot stability is primarily a function of two things,
the directional stability of the vessels and the response time of the
control system. If the vessel has high directional stability, then the
autopilot does not need to be very fast to maintain directional control,
this is commonly referred to as passive stability. If the vessel is
unstable directionally, then it tends to broach when over powered by
following seas (AKA pooped). A fast autopilot can sense the course
deviation and compensate for it before the error reaches the point of no
return. This is referred to as active stability.

Directional Stability: It is true that a full displacement double ended
hull has better directional stability than a semi-displacement hull,
however directional stability is also a function of the length to beam
ratio. Therefore, a semi-displacement vessel can achieve the same level of
directional stability as a full displacement hull, provided it has a
sufficiently higher length to beam ratio (I know thats blasphemy to full
displacement advocates, so I'll probably get hate mail for months). The
problem is modern manufactures have optimized the internal space to slip
fee ratio, which equates to low length to beam ration and high A/B ratios.
These conditions combine to produce the ultimate party boat that is unsafe
beyond the breakwater.

Active Stability: The key to achieving active stability is generally
limited by the quality and rate of the directional sensor. Commonly
autopilots use a fluxgate compass to sense directional changes, which
typically provide 0.5 degree accuracy and up to 10 Hz output. This means
the sensor can not sense a change of less than +/- 0.5 degrees and it
takes 1/10 seconds. In the worst case a vessel could rotate from -.5
degrees to +.5 degrees in 0.1 seconds, i.e. a rate of 10 degrees per
second, without being detected. If the course deviation rate is
increasing, the course deviation could be several degrees before it is
detected. The problem is that the further the boat gets off course the
greater the control moment (rudder force) that is required to correct the
error and return the vessel to its intended course. Many people try to
compensate for marginal stability by increasing the size of the rudder
and/or the thrust of the hydraulic RAM, however more stability margin can
be gained by improving the quality and equally important the rate of
heading data. Larger and more expensive ships rely in inertial rate
sensors (gyro$), but an attractive new option is the Hemisphere Crescent
Vector GPS http://www.csi-wireless.com/products/oem_vectoroem.php. It is a
24 channel GPS receiver with two antennas. When the antennas are mounted 2
meters apart it provides heading (to True North) to within 0.1 degrees at
20 Hz (5 times more accurate than a fluxgate compass and twice the data
rate). It also utilizes carrier phase tracking to provide 2.5 meter
position accuracy 95% (1.25 meters one sigma).

Regards;
Mike Schooley