rms dev ✨

    Computing Contact

    There are two styles of contact. One which reasons about Penetration/Collision/Friction and another which uses a LPM where the rigid bodies are attached via a spring-mass system.

    Hard Contact

    which can be further divided into two classes:

    600
    LCP (pyramidal approx. of friction cone) vs NCP (true friction cone)

    LCP

    Linear Complementarity Problem

    There are different solvers used to solve this, common ones include (iterative or direct):

    • PGS
    • ISOR
    • Danitz

    Issues

    •  iterative methods can lead to the body rotating -> can lead to asymmetric result depending on starting position

    This is often non-differentiable. Nimble makes this differentiable for DART.

    NCP

    Dojo solves this exactly, without approximating the friction cone using a fancy interior-point method.

    cone.gif

    600

    Soft Contact (Spring-Mass Model)

    Note: is there a damping coeff?

    Issues:

    • experiences unphysical interpenetration
    • forces at a distance (i.e., while not in contact)
    • need to tune the contact model parameters