I made the trek over to the engineering library at the University of Washington this weekend and looked up several of the papers about powered soft-landing trajectory optimization that I wasn't able to view online. The papers weren't as useful as I'd hoped; they all tend to assume a flat earth with constant gravity, which makes the problem much simpler. My world is exaggerated (smaller world, faster time) so my powered flight occurs over a larger range of gravity strength and direction.
Unsurprisingly, it turns out that fuel-minimizing trajectories get as close as they can to two impulsive maneuvers: one at the start, and one at the end. One paper I read calls this a “bang-bang” trajectory. Cute!
On my program, I did a few minor things. I put in an engine cutoff at touchdown to keep the rocket from rebounding off the ground. It's hard to let off on the rocket at the exact instant it touches down, so it is nice to have the game do it for you.
I also enhanced my time acceleration. On a “bang-bang” trajectory you spend the middle part just waiting while the rocket coasts. This is boring. Holding down the Tab key speeds up the game clock. Previously it just accelerated time by 25 times. This was fine for low orbits but much too slow for larger orbits, so I decided to make it variable.
My first attempt was to try to keep on-screen speed constant. This meant that the acceleration was proportional to the view radius (which changes as the camera control algorithm zooms in and out), and inversely proportional to the rocket speed. This worked great, except when it didn't. You can probably guess when it didn't: When the rocket speed approached zero. For example, you're sitting on the ground and press Tab. Bang! Infinite speedup. The other bad case is when you're flying straight up; the rocket will slow to a stop at the apex of the trajectory, then start falling back down. If you are accelerating time through the stop, suddenly your rocket is back on the ground, in pieces.
Because of these problems, I have dropped the inverse proportionality with rocket speed; the time acceleration is simply proportional to the view radius. I'm not sure whether this is wholly satisfactory. The camera frames orbits (up to a certain size), so if you are flying a highly elliptical orbit you will zoom around the periapsis at high speed and then dawdle through the apoapsis. Periapsis is a common location for doing burns, and having the time acceleration too high there makes it hard to hit the right spot in the orbit.
I added a game clock and a wall clock to the HUD, and a display of the time multiplier when the Tab key is down.
Finally, I spent some time trying out various possibilities for the camera-framing algorithm when the orbit is very large, or hyperbolic. I don't have anything satisfactory yet. Ideally it should return similar results near the boundary with the orbit-framing regime, so the camera adjusts smoothly from one to the other. It needs to show the rocket at all times, and probably some portion of the unflown trajectory. It's also probably important to have the planet always be in view.
An example I tried: find the periapsis point, and the axis of the orbit. Expand a bounding circle from the periapsis point in the direction of the orbit's axis until it hits the rocket. Unfortunately this suffers from problems when the rocket is near periapsis. It also doesn't show too much ahead of the rocket for elliptical trajectories. On the plus side, it shows almost all of trajectories that are not too eccentric. It's probably my favorite out of the things I've tried so far, though.
A couple of years ago I was at the Game Developer Conference, looking over the entrants for the Independent Games competition. A couple of Digipen students (Kim Swift and another guy whose name escapes me) were there demonstrating their student project, a little game called Narbacular Drop. Valve saw the game, too, and hired the entire team of students to turn it into a shipping product, a process that took about two and a half years. I spent a couple hours this weekend playing it. It's pretty fun up to the point where I'm at (room 18), where it turned maddeningly difficult.