Idea for celestial engine and editor. Important and useful.

This post is mainly directed to Jon Mavor and planetary system physics programmers who work on Planetary Annihilation.

On may 17th 2013 video @ 25:08 John Mavor said that they are trying to use real physics for planet simulation. I am actually physicist from one of the world's top physics schools, who also loves programming and is good at it. I think that with real physics you can make very cool and very original and difficult to duplicate solar systems, that would make game cool and original. Such actually real systems can be very interesting especially when a gravity powers the orbital motion of celestial bodies and when each celestial body interacts gravitationally with other body, for example not only sun but also other planets exert gravitational pull on the planets, and when moon is not only pulled by planet it revolves around but also by sun and other moons etc. Imagine for example planets or stars rotating not one around each other but one around common center of mass. Or imagine moon with moon. It is realistically possible that even moon can have its own moon. If these moons were of comparable mass and close to each other with proper velocities they could revolve around their invisible reduced center of mass, and both rotate around the planet. Real physics would make the game actually ultra cool and fun. For example when 2 big moons rotate each other and go around the planet at the same time, their orbits could be unstable. You cold hit such moon with not to big asteroid just to knock it out of the orbit. Result? The result is real fun, because then it could collide with other moon, fall on the planet or gravitationally pull other moon, so the other moon could fall on the planet too even when it wasn't hit by an asteroid, or its orbit was stable enough, before asteroid hit its partner moon. Real physics provides a lot of room for cool and smart gameplay. Imagine that you want to smash a big moon into planet. You will not have to attach ton of engines to it if the solar system is smartly designed. You could just attach the engine to the small asteroid that would transfer its momentum to the moon by hitting the moon. This could result in just slightly knocking the moon out of its actual orbit that the moon's orbit will change and be less predictable, and maybe in the future planet's gravity will do the rest of the job after the moon hits the planet. Newton's gravitational law is enough. Relativistic gravity effects will complicate things too much, and be mostly insignificant for the most of simulation. But programmers have to remember about momentum transfer and inelastic collisions (kinetic energy between celestial bodies is not conserved because huge part of kinetic energy of impactor is changed into heat, light and deforming itself and a target).

The only big problem with realistic planetary system is that you actually have to forward simulate it to predict its future. You really don't know what's going to happen in the future unless you get there (or accurately simulate into the future). You have to simulate it into the future to make sure it is stable enough. Of course future simulation should be same algorithm as in game simulation with the same step of time , and same parameters. Otherwise simulation can surprisingly give different outcome then gameplay.

Tip for the planetary system programmer: Don't use Euler's method for accurate simulations. Euler's method is not good enough. It will not yield stable (or stable enough) orbits even when they should be stable in real physics world. use Runge-Kutta method. Maybe 4th order Runge-Kutta is going to be good enough but I believe that 16th order Runge-Kutta method should provide excellent quality of real physics simulation especially for the game purposes. It is going to be difficult to provide realistic simulation for long time and difficult to guarantee stable orbits when using Euler's method even when in real world given orbit should be stable.

Celestial engine:

0) should provide some useful realism close to real physics.

1) should provide ability for creating celestial bodies that have elliptical non necessarily circular orbits.

2) should provide possibility of creating orbits in multiple planes.

3) Should simulate pieces of destroyed celestial bodies so they could go into space and collide with other celestial objects, transfer their momenta and destabilize orbits of other objects if these chunks are big enough.


Celestial editor:

0) Should provide all above

1) should provide simulator, that would simulate celestial body motion to see what happens to it in the future (for example that planet doesn't fall on sun, or collide with some huge asteroid for sufficient amount of time). I strongly recommend @ least fourth order Runge-Kutta method of simulation. 16thorder Runge-Kutta should provide excellent physics quality simulation.

2) Should provide distance or velocity vector solver to guarantee closed (for example elliptical) orbit for important celestial objects that should have closed orbits. For example, players are going to start on planet A. When someone builds the solar system in celestial editor, he must make sure that if the planet A is placed certain distance from the sun then it must have certain velocity to achieve closed orbit with given eccentricity. Solver should find such optimal planet initial velocity vector, or optimal distance of the planet from the sun that will guarantee that planet moving at certain velocity will not escape the system nor fall on the sun, but instead it is going to have certain orbit with given eccentricity. The solver should work also with moons or asteroids that go around planets, moons or other objects. Guarantying such solution for complex system may be very difficult, but at least it should help finding stable and closed orbits of some simple systems. Even solver of the simple system + future simulator could be very helpful in design of planetary systems.


Celestial bodies:

1) Should drop shadows on other celestial bodies when between the sun and these other objects.

2) should have darker nights on dark side according to may 17th 2013 video dark side of the planet is not dark enough.

Implementing real physics would have teaching value. Design cool planetary system and see how it behaves.

Re: Idea for celestial engine and editor. Important and usef

Oh, also, for some ideas on how to manage things like rockets doing realistic inter-planetary transfer, I suggest you ask r4m0n, who is very notable for creating the MechJeb mod for Kerbal Space Program.

You can find him at the kerbal space program forums, and he's at the #planetaryannihilation channel at irc.esper.net

Re: Idea for celestial engine and editor. Important and usef

I'd love having some good physic support as well. As with all good ideas though it comes down to: Can it be done without sacrificing another important feature and will Uber be able to stay on schedule. I really hope it works out. I would probably spend days or even weeks building crazy solar systems.

Btw, the 'dark side of the planets isn't that dark to make it easier to see your units. I guess there is ample of ambient light to keep units readable for players. But maybe the value for ambient light could be exposed to the map save file for easy editing?

Re: Idea for celestial engine and editor. Important and usef

Even better would be if it was exposed as a per-user setting, kinda like the usual 'gamma' settings except just for dealing with the 'dark' side of planets.

Mike

Re: Idea for celestial engine and editor. Important and usef

How about just one setting that looks good?

It looks flat right now as it is. What you need is more contrast in the shadow.

Re: Idea for celestial engine and editor. Important and usef

No.

Re: Idea for celestial engine and editor. Important and usef

No.

Re: Idea for celestial engine and editor. Important and usef

There is no perfect setting. There should be diversity.

Re: Idea for celestial engine and editor. Important and usef

Well, the dark side doesn't have to be dark like it is (politely saying) in native African butt! It could be just twice as dark as on may 17th 2013 video. You would see your units well. Besides that units could have some luminescent team colored textures to be well visible on dark side. That would work perfectly especially for someone whose team color is black (just kidding).

Re: Idea for celestial engine and editor. Important and usef

Well I think that Runge-Kutta will speed up the schedule. Model is a little bit more computationally intense than Euler's method, but actually it works better and is more suitable for real world physics. With Euler's method you are most likely to fail real world simulation especially if you want your system to behave realistically on long time scale (for example 4 or 8 hr game). If you simulate cannon projectile that goes from pint A to B on short path and has parabolic orbit/trajectory and lives for about 5s or at most 10s then Euler method can be sufficient for such simulation (for example if you simulate canon projectile from Mavor in SupCom). But if you want to simulate motion of celestial bodies, or projectiles that are fired from planet to planet and where these projectiles don't move on parabolic orbit, where such motion realistically depend of gravitational and kinetic interactions then Euler's method is going to fail especially if you want to simulate celestial body motion and results of their collisions for tens of seconds and especially hours not seconds. When you stick to Euler's method your simulation will fail and you will try to simulate stuff on microsecond timestep scale and you will still probably fail. That's the waste of resources and that's the stuff that may push the schedule.

Besides that, implementing Runge-Kutta method is way more effective. It is more complicated than Euler's method (mathematically), but you don't need to implement new systems, software modules etc. College level graduate with good math skills and understanding of physics should be able to change the entire physics calculation functions from Euler's to Runge-Kutta in matter of two days, and not more than a week. You just change body of the function that calculates new position and velocity vector of the object. That's all you need to do. You do not need to implement physX or Havok or other stuff like that. Runge-Kutta is not such complicated. Not even qarter of it. If you would like to achieve similar results with Euler's method you would have to make time step so small that you would probably waste at least same amount of computing power as for Runge-Kutta.

Re: Idea for celestial engine and editor. Important and usef

i feel complied to chip in here...

I think doing a full physics simulation - in the sense that the system is reevaluated every frame - is over the top complicated and unnecessary. Like you say, on generation every planets and orbital bodies path should be simulated into the future to the point where either a closed Orbit or collision is achieved. Once this is done though planetary bodies should just move on rails - until there is an influence event - such as the fusion engine thrusters from the Gameplay Visualization firing.
Then things get more complicated, but in my opinion you have to consider what the user actually does, if he/she could just switch on the thrusters at will you would have to do a full simulation again. But actually all the user will do is designate a target, and the engine will then provide a trajectory solution and visualize that. Everything further than that would really distract from the gameplay.
If you think about it it's going to get complicated enough as it is. How do you actually visualize a trajectory? Like they did in the Visualization? I hope not.



If you look at the trajectory shown for the KEW impact (let's ignore how it makes no sense from a simulation point of view) it completely disregards planetary movement and rotation and moon movement. For all you know the moon's going to be in the way or you are going to hit the wrong side of the planet once the asteroid's there. And even more important: There's no easy way of visualizing this information.
Really getting a trajectory solution for every spot on a planet that a user may wish to hit will get ridiculously complicated. Especially if you factor in that you really need to convey that information to the user so that they do not get surprised if their asteroid suddenly takes 20 minutes to it's target instead of 10 when all you did was slightly move the designated target on the planets surface... ( They might actually have to cheat there)

So basically what i'm saying is no real time simulation, just event based reevaluation. Otherwise you will have to do a real ugly bunch of prediction whenever you want to generate a KEW event, and they might not even hit where predicted if there is any simplification involved.

Re: Idea for celestial engine and editor. Important and usef

The current implementation uses verlet integration. This provides a good balance of stability and computational simplicity. We have not yet implemented any of the RK methods, but as you mentioned, doing so would be a fairly simple piece of work should we choose to go the route.

As much as I love watching the n-body simulator run, and dream about complex and chaotic star systems and the kinds of emergent fun they would generate, I fear that the lack of predictability would run counter to the primary thrust of an RTS. When N is bigger than 3 or 4 for any localized region of space, really crazy things can happen. At the end of the day, in PA you need to be able to click on an asteroid, click on your enemies base, and have it "just blow up"; for the most part. And to that end, I think we may end up with a much more simplistic physics model where the planets largely run on rails (until they don't). We'll see.

Two things to keep in mind:
A) We're going to play around with it both ways. See what works and what doesn't. Get feedback from the community. And ultimately use the system that seems to be the most fun.

B) One of our guiding design philosophies is to expose as much of the game as possible to the modding community, so that everyone can build/extend/play the flavor of PA that they like best. To this end, even if stock PA doesn't use a true physics simulation for planetary motion, we will try to make sure that the ability for modders to turn on true physics is exposed by the engine.

Re: Idea for celestial engine and editor. Important and usef

You welcome

I agree, there is no need to make full electromagnetism simulations nor general relativity simulations nor quantum mechanics simulations. Nonrelativistic classical mechanics simulations are enough. Relativity can be reserved only for simplest stuff like converting mass to energy and vice versa: E=Mc^2.

I didn't say that. I said only about simulating into the future. There is no guarantee that every object will collide or is going to have closed orbit. Making simulation for example 80h into the game is good enough. If some planet survives simulation for such long time, there is not need of making sure that it has closed orbit. Not many players will survive 80h long matches and most games will be over in such time. How long to simulate into the future? Ask the designer, enter number of minutes or seconds or hours you want to simulate into the future.

You are at least 100% wrong!!! What's the point of making simulation if motion of object is defined and calculable from formula evaluable in a single step? Having simulation when objects are on rails is unnecessary. On the other hand if objects have hardcoded path then there is no need for simulation. Game must be based on exactly the same model as simulation with exactly the same parameters to make sure to yield the same result as simulation. Otherwise simulation is worth nothing. What you wrote doesn't make any sense. When it comes to influence events my point is that such event's should be present at all time -gravity is a constant event. My point is that realistic physics engine gives more options. For example if gravity is no influence event but only fusion engine is, then to smash moon into planet you will have to add huge number of fusion engines to the moon and direct it into the planet. real physics gives more options for smart players. Imagine planetary system with pair of moons orbiting each other and where that pair orbits a planet. In real physics one of these moons pulls other one away from the planet. In the game if there were no real physics, destroying one of them would make other one move like the other one was still there. That's really goofy and primitive. You blow up the planet and the moon will still go around it on rails like the planet was there. But if you have real physics with 2 moons in pair orbiting each other, you kill one of them with asteroid and if pieces of destroyed one are small enough and fly off far away then the other moon is going to be slingshot in space or attracted to planet to collide with it. This all can be achieved with intelligent gameplay available thanks to real physics. In smartly designed planetary system you knock one moon away with small asteroid and lack of that moon will make other moon go on planet. Or you attach thrusters to just one of them and you change path of both moons. You can collide 2 moons with planet by actually adding thrusters to one of them. That's coolness of real physics simulations. If everything was moving on rails other moon wouldn't notice absence or change to the orbit of other moon but also to the planet it orbits. You would have to script everything "what if" to have changes as the effect of changes. When you have good and realistic physics model you have plenty of problems solved once your system works. You just test it, make sure that it works and you can forget about everything else. Physics will solve everything for you. In my opinion, realistic physics model will definitively make game cooler, smarter, and will simplify development.
Removing real physics actually makes things complicated. How moons are going to move when their planet is destroyed? On what orbits? When you have system on rails you have to answer plenty of questions, and you may omit some questions and answers. When you have gravity everywhere and good quite realistic physics model then physics is going to provide answers for you.

. When you have a physics simulation based on good realistic model going all the time you don't have to worry about anything at any time. Physics will do all the work for you. Target designation my work well with realistic physics well too. Computer may try to adjust thrusters in such way to hit the target or you do it manually. The game could use attractor fitting, or solver for making small asteroid from far away hitting particular region of the planet. That's computationally intense but accurate solution doesn't have to be provided in single frame of the game. Game can approximate path of thrusted asteroid and make approximation more accurate or fit parameters precise with each frame of the game or with each game km traveled. That's not much more complex than adjusting turret angle to hit the target.
You can approximately visualize trajectory. You can ask NASA how to do that, or get scientific papers on it. There are definitively some not to expensive algorithms to do that for example patched conic approximation (PCA). They are simple, realistic and general and therefore may be perfect for the game.

There is: Chronocam with forward (future) trajectory simulation -in form of video.

Yes if you want to solve it analytically, but who a hell needs analytic solution. Simulation is sufficient and PCA may be really good enough. You don't need perfect solution right away even through simulation. When your asteroid is far from target it needs approximate solution. You divide entire path into points A, B, C, D, etc and you solve, just for these few points. Once you got to point B, you make more accurate approximation for C, D, etc. And maybe divide remaining path into same number of subsegments. This will make your approximation better and better. When you are close to target you start to worry about making sure you hit the proper point. You don't care about it in a single frame on the beginning. Problem is not so difficult as you think. This is the game and people send stuff to Jupiter, Saturn, Mars, and their moons or even asteroids. People landed on the Moon in late 60's, and the solution how to maneuver stuff between Earth, Moon, and other planets was well solved in real life decades ago. Having good system for the game that even works realistically is a piece of cake.

I don't think there would be such an impact, but even if there was so what? Then asteroid would hit 20 minutes later, and so what? Simulation can take into account only what it knows. If conditions suddenly change then outcome of simulations is going to be different. That's cool and realistic. I don't think impact time would ever shift by 100% especially if you start far away and you have long way to go at not too fast velocity. If you are close and impact time shifts even by 100%, then it is going to be short time anyway because you are close.

I don't think there is any problem with it. Even real life algorithms are simplifications and are really good. There often is room for corrections even in flight in real life. People landed on the moon, and your computer is way more powerful what NASA had in these times. Trajectory to the moon as far as I know was calculated on IBM 702 computer. It could:
I/O :~15kB/s from/to tape, or ~25kB to magnetic drum.
do 4 000 five digit additions/s,
do 833 multiplications/s (5 digits * 5 digits),
do 400 divisions/s (4 digit / 6 digit)

Had RAM: ~10kB with internal transfer rate of 43kB/s.

This was used to calculate trajectory to the moon. Desktop PC is way faster. It could probably do in blink of the eye what that computer could do in hours.

I think you are panicking too much about the problem. PA, has way more complex algorithms implemented than physics simulations I propose.

I want better game. Realistic non-relativistic classical mechanics simulation where gravity holds the system together is a great idea and will make Planetary Annihilation way better than some simplified goofy and boring physics engine like you had in spore.

Re: Idea for celestial engine and editor. Important and usef

Ludicrous stance to take. Either the shadows afford visibility and look good or they don't. It's really not a matter of preference.

Re: Idea for celestial engine and editor. Important and usef

Considering the differing views that should be apparent that there is some element of preference involved.

Frankly, so far as we know right now all the night day stuff is purely visual, in that case there isn't any reason why it couldn't be opened up as a User Setting as I already suggested, it's pretty much win-win, everyone gets what THEY think is best. For example I think it would better if it were a touch darker than what we've see so far, but hey, maybe someone else would find that too dark, or maybe they have some complication with thier eyesight, everyone is a little different.

Mike

Re: Idea for celestial engine and editor. Important and usef

UberGaf: Thanks a lot for your response.

Checking it wouldn't be difficult. Build one simple planetary system (very massive sun, and light planet that should orbit the sun on circular orbit (make perfect setup for this with perfectly adjusted parameters)). Run simulation for several (for example 5 or 100) game years (revolutions of the planet around the sun). Make the planet draw the tail (trajectory). See if trajectory is a perfect circle. The more perfect circle the trajectory is, the better quality of simulation engine.
You can implement RK into Verlet method.

If the engine is not based on analytic time dependent equation of motion then it is not predictable. Predictable system gives equation. You plug in time t (for example trillion years into the future) and immediately you get the solution of the state after the time t (for example trillion years). Otherwise if you have to simulate then its a simulation and the system is not predictable right away.

My point of removing the rails is such that:

what happens when something orbits something that is on rails, and that something gets blown up? You have to script solutions to whats going to happen to the moon that looses its parent planet.

What happens in complex multibody system when you remove some important body. If something is on rails basically it is not going to notice that something important got removed.

When you have constant interactions like gravity, you could affect one object by affecting the other one that strongly gravitationally interacts with object you affected. As I wrote this will make possibility of designing complex star system where such complexity not only looks cool but also can be useful for smart gameplay. As I wrote before -two coupled moons. You slightly destabilize a little bit one of them and both or the other one hits the planet. You could actually knock the stuff around not only by impacts but also by using gravitational interactions or by slightly destabilizing systems in sensitive gravitational equilibrium. I know that realistically such planetary systems are going to be difficult to make but, that's why editor needs solver: I put such big sun there and such small planet here and what should be planets velocity vector so it has elliptical orbit with such eccentricity. Or planet has such velocity and how far it should be from the sun, to have closed orbit? Solver should find such parameters. If it could be applied to forexample coupled or tripled moons going around planets than it would be awesome. Possibilities are endless. For example -one planet stealing moon from another one. Realistically it is possible in real physics. I am not saying simple to set up, but realistic. Imagine having moon base from which you can watch your planet, and suddenly other big planet passes by that has orbit in another plane and steals the moon with your base. Now you can easily colonize that another planet that has different orbit, with different eccentricity. Realistically Orbits of moon can be elliptical from planets perspective. Then at certain times the cost of escaping that planet to the moon may be very cheap and at other time very expensive. When I think about possibilities that real physics can give to such game then I am amazed. I am physicists who loves computers and programming so for me it is double fun to have realistic planetary physics in such game.

If chronocam is not screwed up then what's the problem? Jon Mavor said that the game engine takes snapshot of the state of the game. If that's true than you have new initial conditions for the solar system with each snapshot for each snapshot. Backwards and forward simulations shouldn't screw the system up, and shouldn't mess up simulation (simulation shouldn't mess up a chronocam, and chronocam shouldn't mess up a simulation). I am not sitting in your code, so I don't know what you mean. RK method actually should make simulation more precise and it actually should work with Verlet method.

. Exactly that's what I am waiting for! Simple predictable maps, or complex ones where crazy and cool and funny things can happen. Imagine you jump on the moon with plenty of resources. You don't know that if you don't get out of it in certain amount of time you will get gravitationally slingshot into space because of N body interaction. Someone should be able to design such map, without scripting but with just physics in it. When you escape/jump, where, and if you escape to the moon or another planet at what time should matter a lot. Scripting cool rails can be difficult and time consuming when physics is a script for coolness itself. You just setup initial conditions and let it go.

I am not against it. This can be achievable with real physics where you have planet with asteroid belt around it. I am not against it. My point is when what if someone wants to do something ingenious and awesome on map, where some one wants to use complexity of the system to destabilize part of it. If he does it the smart way then small gravitational pull or thruster push or impact knock (momentum transfer) on merely stable system, could lead to smashing not only to asteroid but also a moon or two planets together. That's way cooler than just "send this asteroid there". Pieces flying off the explosion flying through space and being gravitationally atracted by moons or other planets. You play with kinetic weapons and gravity pulling in the game, and if you do it right and in the right moment you can get of avalanche of consequences, not just one impact, but actually maybe many impacts from pieces of one impact, destabilized systems -planets and moons with bases failing into sun, colliding or being gravitationally slingshot into space etc.. Simple maps maybe will not give such possibilities but complex ones could be full of such possibilities. OK so how you kill a guy that is lonely on a planet far away on the orbit on different plane that has eccentricity for example 0.98? You direct asteroid there. He is quite safe and alone turtle there but on another hand with no place to escape.



Two things to keep in mind:

I am waiting for my new debit card. I plan to buy the game soon with alpha access. Please give us good editor, skirmish and realistic physics. We want to test both ways. At least I would like to try these concepts I wrote about. If realistic physics is rejected, or the other one is rejected then add it as a mod. So one who likes realistic physics could make maps that use it as a mod for the map.

Where can I get documentation of the engine or modding tools with documentation?

Re: Idea for celestial engine and editor. Important and usef

I just wish we could include the electrical interactions as well as the kinetic ones. Magnetic's and electricity are to ignored in celestial physics in the mainstream sciences, when it has much to do with the surface features of our planet as well as mars. Not only would it be awesome, but if you were able to move a object large enough to the target zone the difference in potential between the objects would result in some fairly cataclysmic lightning between them.

Re: Idea for celestial engine and editor. Important and usef

And make realistic thunderstorms and planets throw bolts like vanDeGraaff generators.

Re: Idea for celestial engine and editor. Important and usef

Sokolek, based on your responses here I think you are looking for a different game. Something more like KSP but for planets perhaps?

Regardless the emphasis here isn't on hardcore simulation which seems to be your preference.

Re: Idea for celestial engine and editor. Important and usef

At that point, your really just making Dwarf Fortress with fancier graphics and self replicating machines of war.

As much as I would like a fully realized simulation of celestial bodies, having to check the planets rotation, atmospheric condition ect, for every shot would take some processing power. Considering the number of weapons being shot, and the different calculations needed for every projectile, it's easy to see where some problems might appear.

Having reasonably good, small calculations that don't need to be run every moment should be fine, even if they don't take into count everything they should work fine. Just having the simulation run a while when its making a map and reference where the planets would be ingame is fine until mid-late game where you have asteroid engines moving around.