FORC · Lecture 5

Gravity compensation and inverse dynamics

Simeon Nedelchev

Model-based compensation

M\dot v+c=u+\tau_{\rm passive}

Gravity compensation uses the gravitational bias.

Bias compensation also includes velocity-dependent inertial terms.

a^\star=\ddot q^\star-K_d(v-v^\star)-K_p(q-q^\star)

The reference must specify position, velocity, and acceleration consistently.

u=Ma^\star+c-\tau_{\rm passive}

Exact cancellation assumes an accurate model and sufficient actuation.

For $\tau_{\rm passive}=-bv$ , cancellation contributes $+bv$ .

Leaving physical damping in the loop changes the error equation.

qfrc_bias, qfrc_passive, and mj_fullM expose the required terms.

mj_inverse requires a desired qacc and careful force interpretation.

Reduce the actuator limit.

Does compensation still improve tracking?