Tuning the H5000 pilot the hard way

My last H5000 article was eight shortcuts.

This one is the long road, as described by someone who tunes instruments and pilots for a living.

A few weeks ago a thread on Sailing Anarchy caught my eye: “H5000 pilot made our SunFast 3300 snake downwind – opinions wanted.” Same boat class as half the doublehanded fleet, same symptom we lived with on the J/111 for years. Most replies were what you’d expect.

Then a user called DreamingWet posted a complete method rather than a list of settings.

I asked if I could publish it here. He said yes, credited to his Sailing Anarchy handle and not his name. So everything quoted below is his. The framing and the commentary are mine, and where I’ve shortened or restructured I’ve tried not to change the meaning.

Warning: it’s not a quick read. And it’s not a quick fix.

Start with a reality check

“Tuning a pilot is a LOT, and I mean a LOT of work. It takes a LONG time, minimum a year plus and thousands of nm for the average skipper, to do it right.”

“Randomly changing settings will never produce consistently good results. The solution space is simply too large. There are too many variables and interactions for anyone to reliably guess the right answer from a single observation or sea trial.”

I’ve done exactly that for seven seasons. Rudder Gain on a Tuesday, Counter Rudder on a Thursday, blame the waves on Saturday.

His alternative is six words:

“SEA TRIAL. RECORD. COMPARE. UNDERSTAND. DECIDE. REPEAT.”

Step 0: strip the pilot naked

Disable leeway wind correction. I mentioned this in the last article. He’s blunter about why:

“Entering one value for K for every condition is fundamentally and mathematically 100% wrong. Is one K value better than nothing? 100% not. There are nearly infinite values for K.”

High-end programs run arrays with thousands of K values. The H5000 gives you one. Unless you have a world-class navigator feeding the system, turn it off.

Disable every Expert System. TWS Response, Gust Response, Heel Compensation, Recovery, Auto Response.

“Short of disabling the pilot itself, disable them ALL. You need to understand the basic pilot before adding anything that changes its response.”

AutoTrim off during testing. It’s a fixed condition in his procedure.

Motor Output 100%. Rudder Zero is a baseline, not something you adjust per profile.

Then do the baseline once: dockside commissioning, rudder calibration, compass calibration, initial AutoTune. After that, leave it.

Rudder stall angle

“Under sail if you have to do an all rudder you are in trouble, the rudder cavitates, and all kinds of bad things happen. Know your max arc to port and starboard, know when the rudder starts to stall under different sailing conditions.”

If the ram is well aligned (at half ram travel the rudder is within ±2° of zero), he uses the stall angle plus 4–5% as Max Rudder rather than the mechanical stop. Start with actual max rudder minus 5% and refine later.

“These stall angles have huge ramifications on the pilot dynamics. The significance can’t be understated, but getting this right is the last thing you should do.”

The sea-trial matrix

This is the core of it, and why it takes a year.

Every combination of wind range, sea state, boat speed and Perf mode is its own sea trial. Under engine, so you control boat speed.

“The purpose of each sea trial is to allow the pilot’s PIDs to learn under that specific combination. Do not treat several different combinations as one sea trial, and do not rush from one to another.”

Starting conditions:

• No wind, no waves
• 4 knots with the corresponding sea state
• 8 knots with the corresponding sea state
• 12 knots with the corresponding sea state
• Later: 15 knots, and different sea states within the same wind

For each wind range, run trials at 3, 4, 5, 6 and 8 knots BSP (if the engine gets you there), and max. For each boat speed, a separate trial per Perf mode.

So “no wind, no waves” alone means 3 kt at Perf 1, 2, 3, 4 and 5. Then the same at 4, 5, 6, 8 and max. Then repeat for every wind range.

Why run all five Perf modes when he’s about to tell you to stay away from two of them?

“Because it will teach you a lot, and help you to understand why one should stay out of Perf 5.”

And why not assume 4 and 5 knots give the same numbers?

“The settings between one BSP and another may look similar, but they are not necessarily the same. Internal variables change and are not exposed to you: boat inertia, ROT, etc. That is why every combination is its own sea trial.”

Write it down, then memorise it

Each trial becomes a profile, named so anyone can read it:

ENG-CALM-3KT-P1
ENG-8TWS-5KT-P3
ENG-12TWS-6KT-P4
ENG-20TWS-MAX-P5

Record wind range, sea state, BSP, Perf mode, Rudder Gain, Counter Rudder, Response, Rudder Speed, and notes on how the boat behaved.

“Create the table and then memorize it. Someone should be able to give you one value, and you should be able to fill in the rest without thinking. Have laminated copies on the boat. When you are so tired that you can’t read your watch, wet, cold and hungry, these tables will be your guardrail and safeline to safety.”

Once you have the table and understand why the numbers move the way they do, you’re one fifth of the way there. His estimate, not mine.

One change, one observation

Go back out in conditions matching a profile. Start from the learned values. Change one thing. Observe. Revert.

“For example, in no wind and no waves at 3 knots BSP, change Rudder Gain to the learned value from 5 knots. Observe what happens. Return to the original profile, then do the same with Counter Rudder, Response, Rudder Speed, and PERF mode. One change. One observation. Then return to the learned profile.”

If you can’t feel the difference:

“Pay more attention. The differences are there, trust me on that. If you become really good, you should be able to jump from one SF3300 to another SF3300, with everything identical, and feel the difference. Not one boat is identical to the other. This ‘feeling’ is the heartbeat connecting the pilot, the boat, and you.”

If you still can’t, plot it in Expedition: Rudder Angle vs Heading, Rate of Turn vs Heading, Rate of Turn vs Rudder Angle, BSP vs Rate of Turn, BSP vs Rudder Angle. Then run the trial again with the screen on deck.

“Practice until you can tell what changed by feeling the boat, without first looking at the display.”

Under sail

Under engine, boat speed is commanded. Under sail it’s observed. Wind, sea state, point of sail, sail plan, trim, heel and balance decide it. So the engine table isn’t a lookup for sailing, it’s the reference that shows how settings move with speed and conditions.

His race-day method, with his own caveat that it can make things worse on some boats:

EXPECTED RACING CONDITIONS
→ SEA TRIAL UNDER ENGINE IN THOSE CONDITIONS
   (only if you can get within ~90% of target upwind BSP, otherwise skip)
→ ALLOW THE PILOT PID TO LEARN
→ COMPARE WITH THE PROFILE TABLE
→ UNDERSTAND WHAT CHANGED AND WHY
→ DECIDE WHICH SETTINGS APPLY UNDER SAIL

“Do not take a calm-water profile and assume it is correct in 20 knots.”

Perf 4 and 5

I’ve been recommending Perf 3 as the all-round setting. He goes further:

“If you ever have to use Perf 5 or Perf 4 there is something wrong with your tuning, sail plan or trim. Perf 4 in fast planing boats is OK on some specific boat dynamics. Perf 5 will do nothing but amplify the problems caused by your bad tuning.”

On the SF3300s he’s worked with: never Perf 4, even in 35+ knots. The price was a lot of trimming. Without it, “sooner than later the pilot broached the boat.”

Why the H5000 struggles on modern boats at all:

“The entire H5K pilot mathematical model was designed when boats were entirely different: 2–3 times slower and heavier, with much more longitudinal stability due to the hull shapes, completely different sail plans, CoF, ram speeds, and so on. The limitations are baked in, and have been baked in for decades. The tuning job is to make believe that your boat dynamics are somewhat similar to those types of boats while being effective in your boat.”

You’re not tuning the pilot to your boat. You’re translating your boat into something the pilot understands.

Damping and latency

Starting values for an SF3300: 4 or 5 for wind speed, wind angle, boat speed and heading. Heading is the authoritative one.

“Dampening is a Kalman filter that happens in real time, it adds a few microseconds. Regardless of the damping factor, measuring and refresh rates are 100% unchanged.”

“The damping factor works exactly like a shock absorber. Too hard your teeth fall off, too soft you end up tearing up the suspension. Since the entire pilot loop is behind, you do not want to be correcting for every state change. The goal with dampening is for the pilot to miss temporary wind shifts and heading dynamic changes and only react to longer term changes. Your job is to trim into these temporary shifts.”

“The loop is behind” is the key. His numbers: with the H5000 in the loop, the time from measurement to the rudder command being executed is 1.5–2 seconds. Nothing in the settings changes that. Exocet and MadBrain run that same loop in around 12 milliseconds, and with the rams they use, measurement to the boat actually responding is about half a second. On top of that their models run 2–4 seconds ahead, so in his words they can “see what is coming.”

Every correction the H5000 makes is to something that already happened. That’s why it chases, and why damping has to do the filtering the algorithm can’t.

Heel Compensation is not a heel overlay

This came up after I compared the H5000’s Heel Compensation to the heel overlays in Exocet and MadBrain. He was clear that they’re not the same thing:

“In the H5000, heel compensation acts directly on the rudder. In Exocet/MadBrain, heel is an input into one of the three main PIDs, which then feeds the control model. The PIDs are therefore part of the inputs to the control architecture rather than being the complete control themselves.”

“So, in simplified terms: Exocet/MadBrain, PIDs feed the control models. H5000, the PIDs are effectively the control loops. That difference becomes particularly important when multiple overlays are active.”

Which explains something that had puzzled me. Pixel and Madintec both tell you to turn everything on:

“The philosophy behind Exocet/MadBrain is essentially: Please! Use all the overlays. The system is designed so that the overlays can operate together to improve the boat’s trajectory. Running them simultaneously is not something we try to avoid; it is the recommended configuration.”

“That is very different from the H5000, where I would not recommend running multiple competing overlays simultaneously. The control loops can end up working against each other rather than cooperating.”

Same words on the menu, opposite advice. It’s not that one vendor is cautious and the other reckless. The architectures are different, and the advice follows from that.

Cruising Speed

“Cruising Speed is a set point at which all the standard pilot setup parameters are used at 100% of the input value. In your boat I would set it at 6 knots, keep it there until you go through the calibration process. Eventually you will change it according to boat speed, behaviour and Perf mode.”

He’s “all but certain” it acts on the proportional term. On a 10-metre boat, 6 and leave it until the matrix is done.

Where the motion sensor sits

He ran a kinematic model on two real SF3300 ORC certificates to quantify this. The short version:

Gyros measure the same angular rate anywhere on a rigid boat. But a sensor mounted away from the centre of rotation also picks up extra linear velocity and acceleration from the lever arm. The B&G 3D Motion only knows mast height, so it can’t compensate. It treats sensor motion as boat motion, and that goes straight into the wind correction.

His example: sensor 3 m aft, 1 m to port, 1 m above the centre of rotation, SF3300 in 15 knots and 50–70 cm waves. Extra velocity at the sensor almost 3 knots, extra acceleration nearly 0.5 g. Wind errors up to 11° AWA deep downwind, up to 6° TWA, about 1.5 knots wind speed. Dynamic errors, not offsets, so you can’t calibrate them out.

“You can buy an extremely accurate and expensive motion sensor, but if you install it several meters away from the CoR and the instrument cannot mathematically compensate for that displacement, the additional linear motion introduced by its location does not simply disappear. The sensor can be accurately measuring the motion of the wrong point on the boat.”

His rule: 3D Motion and P9 within 0.3 m of the centre of rotation, ideally x = 0, y = 0, z as close to 0 as the boat allows.

“If your P9 and 3D motion sensor are not as close as physically possible to the CoR, either you move them or please disable motion corrections. Otherwise congratulations, you spent a thousand euros to render your wind solution worse than bad and make your pilot more erratic.”

Two more things. The 3D Motion outputs pitch and roll rates but not yaw, so by his definition it “is not really a 3D motion sensor.” And the P9 has a firmware bug where it stops sending on the bus for seconds at a time, “up to 10 or more.” Update it.

Expert Systems, last

“I understand that sounds cool, but these overlays in general do more harm than good. This is 100% specific to the H5000. Pilots like the Exocet, MadBrain and NKE implement these overlays almost perfectly.”

When the base profiles feel natural, add one at a time: Heel Compensation, Gust Response, TWS Response, Recovery, Auto Response.

His closing line

“90% of skippers give up way too early. 5% give up when the learning is just enough. 3% do not give up, but they never move into the final 2%. The final 2% never stop testing, experiencing, and being curious. These are the skippers who win.”

What I’m doing with this

I won’t run the full matrix this winter. Three things are going into the BLUR plan though:

1. Measure where the 3D Motion and P9 actually sit relative to the centre of rotation on the J/99. If they’re not close, motion correction goes off until they are.
2. Build the engine table for the three conditions we race in most. Flat water, 8 knots, 12 knots. Perf 1 to 3. Laminate it.
3. Damping at 4–5 across the board, and stop treating it as a knob to turn when the pilot has a bad leg.

If you’ve run something like this on your own boat, or think he’s wrong about any of it, I’d like to hear from you.

All quoted material is from DreamingWet on Sailing Anarchy, used with permission and credited by his handle at his request. Any errors in how I’ve condensed or framed his posts are mine.