By Javier Rollon · 2021-12-01
I've watched people fly my CRJ-200 for years. Thousands of YouTube videos, forum posts, support requests. And the number one source of confusion — by a massive margin — is the autopilot. Not how to turn it on. How it actually works. What it's doing behind the scenes. Why it does things that seem counterintuitive until you understand the logic.
So here's the explanation I wish existed when I started sim flying.
First misconception: the autopilot doesn't fly the airplane. It manipulates the control surfaces to maintain whatever parameters you've told it to maintain. Big difference. If you tell the autopilot to hold 250 knots, it will pitch the nose down as far as necessary to achieve 250 knots — including straight into the ground if you're already at 250 knots in level flight and you add power. The autopilot doesn't know about terrain. It doesn't know about common sense. It follows commands literally.
This is why real pilots treat the autopilot as a tool, not a replacement. You manage it. You supervise it. You anticipate what it's going to do before it does it. When I'm testing autopilot behavior in my aircraft, I deliberately set up scenarios where the autopilot should do something stupid — because if my simulation lets it do something the real system wouldn't, I've got a bug.
The autopilot controls three things: pitch (nose up/down), roll (wings level/bank), and optionally yaw (rudder). Different modes control different axes.
Heading mode controls roll. You set a heading, the autopilot banks the aircraft to turn toward it. That's all it does — it doesn't touch pitch. So if you're in heading mode and climbing, the aircraft will turn AND climb simultaneously. Some new simmers expect the autopilot to level off when it reaches the target heading. It won't. It was never told to manage altitude.
Altitude hold controls pitch. It adjusts the elevator to maintain a specific altitude. If you hit turbulence and the aircraft gets pushed up 200 feet, the autopilot pitches down to return to target. In the Jetstream 32, this response is quite aggressive — the real aircraft's autopilot is known for firm corrections. I modeled that deliberately, and I get complaints about it. But that's how the real machine behaves. Simulation isn't about comfort.
This is where most people get confused. VS (vertical speed) mode tells the autopilot "climb or descend at X feet per minute." The autopilot adjusts pitch to maintain that rate. But here's the thing — it doesn't manage airspeed. If you set 1500 fpm climb and don't add enough power, the aircraft slows down. Keep climbing at 1500 fpm without power and you'll eventually stall. The autopilot will pitch up harder and harder trying to maintain the rate, bleeding off speed until the stick shaker goes off.
Speed mode (IAS or MACH hold) does the opposite. It adjusts pitch to maintain a target airspeed. Climbing? The autopilot pitches for the speed you set and the vertical speed becomes whatever it is. This is generally safer because you won't inadvertently stall, but it means your climb rate depends entirely on your power setting.
Real airline procedures use a combination. Climb at V/S until reaching a target speed, then switch to speed mode. The FMS in the CRJ-200 automates much of this — the pilot sets up the profile, and the autoflight system sequences through modes automatically. Simulating that sequencing logic took me months. The edge cases are where things get interesting.
ILS approach is the one that generates the most support emails. "My autopilot won't capture the glideslope." Usually it's because the aircraft isn't in the right mode, at the right altitude, at the right speed, at the right distance.
For glideslope capture to work, you need approach mode armed, the localizer captured (or being captured), and the aircraft at or slightly above glideslope. If you're below the glideslope, most autopilots won't capture — they're designed to intercept from above to avoid terrain. If you're too fast, the autopilot can't pitch down enough to follow the glideslope because increasing the descent rate at high speed risks exceeding structural limits.
In the Citation X, I modeled the glideslope capture envelope pretty strictly. The aircraft has to be configured correctly — gear down, flaps set, speed in the right range — before the autopilot will even attempt capture. People think it's a bug. It's actually the most realistic behavior in the aircraft.
The real system is both simpler and more complex than people expect. Simpler because the basic logic is just PID controllers — proportional, integral, derivative feedback loops that adjust control surfaces based on the error between current state and target state. More complex because the edge cases are infinite. What happens when two modes conflict? When the pilot inputs contradict the autopilot command? When sensor data is unreliable?
After ten years of modeling these systems for X-Plane, my advice to sim pilots is simple: understand what mode you're in, understand what that mode controls, and understand what it doesn't control. The autopilot is doing exactly what you told it to do. If the result surprises you, the problem isn't the autopilot — it's the instruction you gave it.
Javier Rollon is the developer behind JRollon Planes, creating aircraft add-ons for X-Plane since 2010. Follow on Twitter.