So I have been wrapping my head around how I would perform response upon collision in my physical actors. So far I have nailed down my dynamic actors into boxes and their composites. And I have no idea what computational/data design I should use in order to have proper collision response. I have following description of my actors:

- velocity vector

- innertia tensor position

- rotation velocity quaternion

I have no idea on how to implement correct friction, movement initiation, tendence, forces from collision points. In fact I do not have even a concept of the theory.

Are responses in physical engines really procedural or they often fall back to some explicit situation describers to simulate real world AAA behaviour?

The standard (for rigid bodies) is to form a system of linear equations and solve them. This can come down to finding an efficient way to invert a conceptually large matrix. Erin Catto's 2005 paper "Iterative Dynamics with Temporal Coherence" (found on the Box2D's google download page) describes mathematics behind this.

Usually to solve a collision a point of contact is required, along with a unit vector describing the axis of least separation, and a scalar describing overlap. Usually the narrow phase collects this information, or in other words, you can get this info during collision detection.

Box2D *Lite* (the Lite part is important) is usually the best place to get started, and can be found at Box2D's website.

Pre-scripted physics sequences are also used. For characters other forms of animation are used too, like forward and inverse kinematics.