I am trying to achieve a spaceship that flies through space and the pilot can get out of their seat and walk around the ship and place physics objects in it while the ship is moving, rotating, accelerating, and decelerating


  1. Subnautica has achieved this to some degree, in their cyclops submarine, you can walk around it as it moves... However, this spaceship will be going quite fast, doing flips and rotating. I am not sure about other space games because I haven't played them.

  2. Sea of thieve's pirate ship mechanic in those boats where you walk around on board


The main issue I have is just making all the collisions work inside the ship, maybe there is an alternative... I was thinking of making a remote cockpit but that wouldn't be that flexible.


I would much appreciate any solutions or helpful resources, directions, or information to help achieve "using rigid bodies inside of rigid bodies, the children being a first-person player controller (preferably physics-based) and random physics objects."

Thanks in advance

  • 4
    \$\begingroup\$ Reference frames are your friend. Intertial, rotational, whatever. Calculate within the reference frame of the ship. If the ship is not accelerating, but only moving very fast, the ship is standing still and without acceleration in its local inertial frame. If it accelerates, it is still standing still in its local reference frame, but everything in it is accelerating in the opposite direction. \$\endgroup\$
    – Polygnome
    May 23 at 16:25
  • 1
    \$\begingroup\$ What @Polygnome said. Fun fact - constant acceleration is equivalent to constant gravitational field. \$\endgroup\$ May 23 at 17:32

2 Answers 2


The ship stays in place and everything moves. Mostly. In this answer I present three ideas of how to go about it, and I'm referencing the games I'm basing them on.

I could not find out what approach they use for Subnautica. However, considering some of the physics glitches I've seen (note: I have not played Subnautica), I'm guessing they don't do much special about it. Except that when you are moving inside the submarine, the submarine is either not moving or moving slow enough that it usually is not a problem.

A situation where we see something similar is elevators. Traditionally games where there is a fight in an elevator will actually have the elevator static, and move the environment (for example using vertex shaders).

A relatively recent example is the game Prey (2017) that has an scripted ambush in an elevator, but it was too glitchy for combat. So for the ambush, they made an stationary replica of the elevator and teleport the player to there, hiding the teleport in the light flicker.

Quotes from "The Making of PREY - Documentary" by NoClip:

Stevan Hird: There was definitively an elevator that was a challenge. It went from being an elevator to being an elevator with combat experience in it. And trying to combine that with a lot of complex physics and some insane player power made with it.


You have a moving player on top of another moving object. And what happens when it wants to move, you know, the platforms wants to move at a certain speed, and gravity wants to shove something through it. Or it moves it from underneath you too quickly, and you don't want the player to fall. So you end up in this weird situation of trying to glue things to elevators so that physics doesn't go crazy.


Anthony Huso: The ambush, specifically, where you have that fight is not in the elevator. It's a fake second elevator. So when we do the power flickers, we're teleporting you into a fake elevator. That's where your combat takes place. And then once that thing is dead (…) we send you back to the other elevator and start it back.

Idea 1: When the player enters the ship, you teleport it to an stationary ship. Looking outside would require a camera trick: position a camera to look from the appropriate position in the world and render that. Similar to what some games do for mirrors or portals. which you could archive rendering to textures or using the stencil buffer, or clearing the depth buffer.

Some space games would usually do that for other reasons (namely to avoid floating point errors that scale in magnitude as objects move away from the origin). An example of this is Outer Wilds.

Quotes from "The Making of Outer Wilds - Documentary" by NoClip:

Logan Ver Hoef: Each planet is a rigid body, like a physically simulated object moving through space, affected by forces. And so is everything else, like everything in the game is.


Alex Beachum: The way with video games you got to keep the camera near the center of the world's coordinate space. Or you start running into floating point error and everything starts kind of jittering. And so we have to keep the player near the mathematical origin of the world. And the way we do that in this game (…) every time we apply a force to the player, we just apply an opposite force to every physics object in the world that's currently simulating. And it doesn't actually really do anything to performance because we're already doing that. We're already applying forces to every object, because everything is moving. It just another number on top of numbers. So when you jump in Outer Wilds, technically every planet's jumping out from under you, and you're more or less not moving.

Idea 2: everything moves, except whatever object the player is standing on. This means that when the player is on the ship, the ship is actually stationary, and thus moving around inside of it not a big deal.

For a more interesting example, take Portal 2. This games has an intro sequence where the room is moving. For this the player is actually moving a controller in a separate room, and then the game applies impulses to the player character to mimic what the character controller is doing, but inside the moving room.

Quote from Portal 2 in game developer commentary:

We solved the problem of trying to compute the player within this highly dynamic environment by putting them in a virtual room that has all the base shapes of the rendered container, but is simply used to compute player navigation. (It's hidden somewhere else in the map). The viewpoint of the player is then parented to the coarse simulation transform, resulting in the final rendered frame. At the end of the ride the player is teleported into the actual game space.

And yes, Sea of Thieves had to deal with that too. I believe they do physics in a replica of the ship which is stationary somewhere beneath the sea, but you see the meshes on top of the ship. At least we know they do that for navigation…

Quote from "How Megalodon, Kraken and Skeleton Ships Haunt the Sea of Thieves | AI and Games":

The big secret is the navigation mesh for the ships isn't actually moving. It's baked deep into the ocean. The navigation mesh for the entire ship is baked into the deep down on the ocean floor and each skeleton is plotting their movement on that nav mesh. This nav mesh doesn't move with the ship and stays at a fixed point. Meanwhile each coordinate on the navitation mesh is transposed from the ocean floor into the ships deck and hull respectively. This has to factor the current orientation of the ship with regards to the oceans surface. Meaning the navigation paths need to follow the actual rotation of the ship as it crashes through a wave.

Idea 3: Combining what they did for Portal 2 and Sea of Thieves we would have a physical environment that does not move where all the collisions and navigation is resolved. However, what we actually see are meshes transformed into the ship space imitating what is happening in the physical environment.

All the ideas presented here have the ship actually stationary so they don't cause trouble to the physics engine. However, I guess you want the motion of the ship to have some effect inside. In that case you might also want to change the gravity strength and direction in the physical environment, so they match the combination of the acceleration of the ship and the gravity it is under.

  • 2
    \$\begingroup\$ This sounds like a great idea! If you're doing something like Outer Wilds' "when the player moves, actually everything else moves the other way", it makes it especially easy. For example: spaceship thrusts forward, and instead every object in space is pushed the other way, except for anything inside the ship. You then have a static frame of reference inside the ship, and can add on artificial gravity, etc. as needed. A nice addition as well as just adding gravity to match thrust, is adding impulses to things inside the ship for impacts too, like the bridge on Star Trek. \$\endgroup\$ May 23 at 1:14

The main issues here are, roughly:

  1. Floating-point imprecision and catastrophic cancellation. An ordinary (single-precision) float has 23 bits of mantissa. Say you're moving at 1.0 m/s, and you add 1mm/s. You end up moving at 1.00100004673 m/s after float rounding, which is close enough. But say you're in a ship moving at 1000 km/s (1 million m/s) and you do the same thing. You end up moving at... 1.0 m/s relative to the ship. The 1mm/s got dropped entirely! This can lead to movement jitter, physics instabilities, clipping, etc.
  2. Figuring out what frames of reference you should use in the first place. Easy in relatively 'static' configurations (e.g. player in ship in world); far more difficult in e.g. a sandbox builder.
  3. Figuring out which frame of reference to use when targeting objects to build / place / etc. This is more of a problem if you're on something like a rail car where you have multiple objects moving close to each other (like the ground and rail car). In some situations you can end up in quasi-ambiguous cases where it's non-trivial to figure out if the object you're putting down should be attached to the rail car or the ground. (Preferring the result that is closest to the player's speed or frame of reference can be a sane default here).
  4. Many game engines aren't really designed for this sort of thing, and e.g. assume that the entire game world fits within a float.

There are a couple of ways of dealing with this:

  1. 'Simply' ensure that you have enough bits of precision to deal with the loss of precision due to cancellation (e.g. by limiting relative speed scales, or by moving to double).
  2. Use synthesized quad (or higher) precision (always, or just where necessary). Almost never a good idea, but still worth mentioning.
  3. Use double-double arithmetic (or extensions thereof) (always, or just where necessary). This is similar, but not identical to, 2.
  4. Use fixed-point, sized to accommodate the worst-case. This is a good idea anyways for lockstep multiplayer, but has a bunch of problems and caveats (a lot of processors (and GPUs!) pour a lot of die area into floating-point processing, fixed-point library and compiler support tends to be worse than floating-point support, etc). (I've been half-jokingly considering using 256-bit integers for everything, in Planck units. It's a terrible idea for many reasons, but it's still amusing to think about.)
  5. Do calculations relative to a local frame of reference, essentially 'moving the world instead of the player'. This doesn't solve the issue in the general case, but can often mask the issue (errors tend to mainly occur far from the player) while being relatively straightforward. Beware that this can cause problems for multiplayer if players aren't close enough together to share said local frame of reference.
  6. Design your physics engine to handle (and do calculations in) nested coordinate systems. This is powerful, but can be... complex.

Also be aware that many of these issues can also affect rendering, not just physics.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .