How are the constraints enforced each time step?

[lijas]

Are joints/constraints enforced in pinocchio in some way? What keeps them from "drifting" further and further away as the simulation progresses. Looking at for example the pendelum example: examples/simulation-pendulum.py

t = 0.0
q = pin.randomConfiguration(model)
v = np.zeros(model.nv)
tau_control = np.zeros(model.nv)
damping_value = 0.1
for k in range(N):
    tic = time.time()
    tau_control = -damping_value * v  # small damping
    a = pin.aba(model, data_sim, q, v, tau_control)  # Forward dynamics

    # Semi-explicit integration
    v += a * dt
    q = pin.integrate(model, q, v * dt)  # Configuration integration

    viz.display(q)
    toc = time.time()
    ellapsed = toc - tic

    dt_sleep = max(0, dt - (ellapsed))
    time.sleep(dt_sleep)
    t += dt

Is there some built in constarint enforcement in pip.aba?
I am currently using a rigid body solver where I run in to problems of the constarints "drifting", and pinocchio looks like a interesting alternative.

[jcarpent]

We follow the algorithmic derivations of rigid-body dynamic algorithms (by Roy Featherstone) that account for joint constraints explicitly in the evaluation of forward or inverse dynamics. More precisely, we use the minimal coordinate formulation. On the contrary, if you have been using maximal coordinates, you will then have to add constraints between connected bodies to enforce the constraint: this will be achieved at the price of numerical errors.

@lijas What is your current rigid body solver?