Adjusting velocity of virtual character controller after collision

[zopsicle]

The virtual character controller implements sliding behavior and this works well. But when the character stops colliding, it will continue moving in the velocity last configured, instead of the velocity as adjusted by the sliding behavior, that would be parallel to the surface slid against. This image visualizes the actual behavior (left) and desired behavior (right) when the character is accelerated northeast at the beginning and its velocity is not modified after.

Similarly, when jumping and hitting the head against a ceiling, the character sticks to the ceiling until gravity takes over.

I managed to solve this by performing the following procedure at the end of my character movement code:

if (jphCharacter->GetGroundState()
        == JPH::CharacterVirtual::EGroundState::OnSteepGround) {
    for (auto& jphContact : jphCharacter->GetActiveContacts()) {
        if (!jphContact.mHadCollision)
            continue;
        linearVelocity = RejectFromNormalized(linearVelocity, jphContact.mContactNormal);
    }
}

Vec3 ProjectOntoNormalized(Vec3 lhs, Vec3 rhs)
    { return rhs * Dot(lhs, rhs); }

Vec3 RejectFromNormalized(Vec3 lhs, Vec3 rhs)
    { return lhs - ProjectOntoNormalized(lhs, rhs); }

but maybe there is a built-in way to obtain the velocity as adjusted by sliding and collisions?

[jrouwe]

CharacterVirtual doesn't store this information, but what I'm using is simply:

Vec3 old_position = character->GetPosition();
character->Update(...)
Vec3 new_position = character->GetPosition();
Vec3 velocity = (new_position - old_position) / delta_time;

To calculate the effective velocity during the step.

[zopsicle]

😄 That's embarrassingly obvious! Thanks!

[zopsicle]

It seems that this approach does not work well with stair walking. If the character walks up a staircase at even a slight horizontal angle, it gets accelerated in that direction. I think this happens because stair walking does not aim to preserve the input displacement (delta_time * velocity), which makes sense since the actual displacement is in part dictated by the stairs’ geometry.

Maybe I should use Update, StickToFloor, and WalkStairs separately instead of ExtendedUpdate. Then I can measure the average velocity around Update, ignoring WalkStairs.

[jrouwe]

Yes, that would probably work. Another option would be for me to change the code to:

https://github.com/jrouwe/JoltPhysics/tree/feature/adjust_velocity

(warning totally untested)

Would that solve your problem?