TheGatewayOfRealities
patreon
The TGOR Movement System
🕑 Added 2019-04-10 21:59:18 +0000 UTC
Hi, it’s Hopfel writing for a change! I will try to also contribute to this Patreon page from now on, giving an insight into our base systems implemented in C++.
Our game needs to accommodate for multiple species of different sizes, body makeup and movement capabilities. We’ve opted to implement our own physics based movement to fit the resulting requirements. In this series of articles I will give a technical overview over what we ended up implementing for our game.
The Unreal Engine comes with a powerful, highly optimised built-in movement system (See UCharacterMovementComponent) which I will often compare our system to.
The pill
Like the Unreal default, our movement system relies on a capsule for collision. A capsule is defined by a radius and half-height, spanning a pill-like volume. Due to its elongated shape it can sufficiently approximate the shape of a character while still being very efficient for collision detection. Due to its hemispheres at each end it also reliably computes the surface direction (normal) on collisions with the floor and ceiling (where for example a box would often report undesired results).
Where our system differs from the Unreal default is that the capsule does not engulf the whole character but only their minimal shape (meaning, the smallest volume the character can reasonably be compressed into without clipping or taking damage). While the Unreal default movement needs to change the capsule size (e.g. on crouching), this allows our capsule to remain constant. This constant shape mitigates a lot of problems connected to getting stuck when un-crouching and allows us to physically simulate the core as a rigid body.
Notice how this means the capsule needs to be hovering in mid-air for movement on the ground. This is achieved with a “mass on a spring”-like system, making crouching a continuous activity that can model recovery after landing or running into a steep slope.
Physics simulation
The movement core is always physically simulated using Symplectic Euler Integration. Physics simulation allows us to accurately define external forces like gravity, drag and buoyancy which depend on character properties like mass, volume and surface area which differ for each species and can even change during gameplay. For example, heavier species (or characters that hold heavy objects) will not just be slower like it is usually implemented in games, they will properly feel heavier.
(60kg on the left, 120kg in the middle, 180kg to the right)
These properties are partially inherent to the species. Heaviness is one example, but also imagine slim lizard characters being able to slip through water faster than chubby dragons due to their smaller surface area. Opposite to that, chubby dragons will be harder to push off a platform. All of these properties are not explicitly defined on these characters but are an outcome of their physical properties. If the slim lizard suddenly carries a giant flat board under water they become as slow as the chubby dragon. If the chubby dragon is tied to helium balloons they become easier to push off platforms.
(characters with higher surface area experience more drag)
All of this becomes very exciting once you add physics volumes to the mix that also have properties such as density and gravity. Imagine micro-planets with spherical gravity, similar to Super Mario Galaxy. Add to that influences like centrifugal forces and acceleration of your frame of reference. This system lends itself to creating gravity puzzles and weird world settings (for example a planet spinning so fast the centrifugal force on the character noticeably counteracts gravity).
(Centrifugal and acceleration forces make the character “lean” towards artificial gravity)
For all of this to make sense we need to be able to rotate the characters. Very often (which is also the case for the Unreal default movement system) gravity is always applied from top down and the only rotation is yaw (around the up-axis), but since we wanted to rotate an object around all axis we needed to look into angular momentum and torque.
Angular momentum and torque are properly implemented with the bounding box approximating the character’s mass distribution. The inertia tensor makes it so certain axis of rotation are harder to rotate than others. This will especially come into play when swimming or flying, where rolling goes faster than tilting.
Drag
Especially for fluids but also for high velocities (e.g. when flying) drag becomes very important. It dampens movements in viscous fluids and gives characters a natural terminal velocity. Drag is computed with the current physics volume density, the character’s velocity and an approximation of the character’s surface area in movement direction. Here we cannot only use the boundary box for approximation since most characters have a relatively sparse surface. To get a reasonable guess we only consider the total surface area of each axis (stored as percentage for each bounding box face) and average over the desired movement direction to get the final surface area.
(counting white pixels and compare with total amount of pixels)
(comparison in differently viscous fluids)
Buoyancy
We have planned very light and very heavy creatures of very different volumes. Here it makes sense that those react differently to fluids. For example, chubby dragons might not be able to dive down because of their low mass/volume ratio. Very small and heavy characters might not be able to swim at all but instead walk underwater (or for example by holding something very heavy like a metal ball). All of these effects are about buoyancy. Buoyancy measures the displacement of matter in your surroundings. Since the surrounding density and the character’s volume is known, the total buoyancy can be computed accurately. We do also account for depth since characters usually get compressed the deeper they dive. They might even start falling and be able to walk on the ocean floor (and possibly start taking damage without protection due to their organs being highly compressed).
(air to the left, very thick atmosphere in the middle, water to the right)
Problems
There are a lot of exciting features already implemented in TGOR that use physics simulation (as outlined above), and there are endless possibilities to add to that. There are however serious problems with this approach (some intrinsic to movement systems in general), requiring a balance to be struck between features, usability and visuals.
- The character needs to be animated.
If movement features require adding dozens of new animations the workload becomes unrealistic. Additionally the animation system suffers under the added complexity and require Ragdoll Physics or Inverse Kinematics to not be overwhelming. Unfortunately these solutions tend to result in visual glitches.
Visual glitch using our custom IK system that adapts to step-height.
- Movement needs to be replicated over network.
The default Unreal movement system, while quite antiquated, has been refined and optimised towards competitive shooters by Epic Games for decades. Expecting to write a movement system that can compete with that in all regards is wishful thinking for a team our size. The main hurdle is making the movement system cheat-proof (preventing flying/teleportation/speed hacks) without introducing rubberbanding. Luckily with this project we don’t have competitive shooter requirements and can cut corners within reason. We are for example thinking about giving hosts the ability to turn off cheat detection altogether when playing with friends to reduce lag.
(Naive movement replication with 90% synthetic package loss, would be unplayable for people with e.g. mobile connection.)
- The simulation needs to be numerically stable.
All physics simulations that use numerical methods like Symplectic Euler need to cope with numerical instabilities. One caveat is computers running at different frame rates, making it so that the simulation can run completely normal with 60 fps but show major simulation artifacts up to the character spazzing out on frame rates below 10 fps. This problem is magnified by the system running in multiplayer where strong velocity changes are induced from server updates. People familiar with earlier versions of Space Engineersknow the (literal) pain of being thrown about in a spaceship on network corrections. These problems are all very solvable but very difficult to test for and hardly reproducible, since specific errors can only appear in certain situations given specific frame rates and network situations.
(Comparison between 30 fps and 5 fps)
What’s next?
There are a lot of internal systems we’ve so far kept from writing Patreon posts about since they tend to be technical in nature. As the game is coming together however these systems become visible and progress can be shown.
More is to come soon, let us know what you think!
Cheers,
Hopfel
Comments
Furry Lover
I looks like you are doing good ! ! I like how you keep us updated ! Keep this wheel rolling ! Love ya !
Mud Dog
This outstanding!
If you keep going like this the game will be so realistic that it may become real. Which would be a good thing. Next your going to tell me that you created life when trying to make an npc act realisticly. Im beond impressed with your work.
Jonathan Brown
looks like it's coming along well
