Bumpy RigidBody2D collision at TileMap seams

[monkeez]

Godot version

v4.0.alpha15.official [432b25d]

System information

Windows 11

Issue description

When a RigidBody2D moves across the point where two tiles in a TileMap meet, the collision becomes unpredictable. This makes it very difficult to make a physics based game such as golf without having unpredictable behaviour which would annoy the player.

Rolling on the ground the ball can bounce randomly:

bumpy01.mp4

When landing the ball can stop dead for no reason:

bumpy02.mp4

Steps to reproduce

1. Create a scene with a TileMap floor (add collision shapes to the tiles in the tileset)
2. Add a RigidBody2D with a circular collision
3. Add a function for hitting the RigidBody2D around

Minimal reproduction project

tilemap_collision.zip

[Calinou]

Duplicate of #32138.

[monkeez]

Would the current workaround be writing my own collision handling with move_and_collide on a CharacterBody2D node?

[Calinou]

Would the current workaround be writing my own collision handling with move_and_collide on a CharacterBody2D node?

It may not fix the issue, as collision results will still return unreliable normals. I think the only 100% robust solution is to write your own physics in GDScript (or C#), using AABBs which are often sufficient for 2D arcade games.

Instead of using one collision shape per tile, generating a 2D convex hull for the whole tilemap and using this for collision is also an option. Maybe using a concave collision shape could work, so you don't have to write your own physics. Generating the shape based on the entire tilemap probably won't be trivial though.

[monkeez]

Using the convex and concave hulls might work for me, but you're right, it won't be trivial.

Is this issue something that could feasibly be fixed? Or is it something core to the physics that cannot be altered?

[Calinou]

Is this issue something that could feasibly be fixed? Or is it something core to the physics that cannot be altered?

The issue can be fixed in core, but it's not easy to figure out and fix without introducing regressions elsewhere. We have a general lack of physics contributors, so any help is appreciated here 🙂