We estimate the state a noisy robot arm and underactuated hand using anImplicit Manifold Particle Filter (MPF) informed by touch sensors. As the robottouches the world, its state space collapses to a contact manifold that werepresent implicitly using a signed distance field. This allows us to extendthe MPF to higher (six or more) dimensional state spaces. Earlier work (whichexplicitly represents the contact manifold) only shows the MPF in two or threedimensions. Through a series of experiments, we show that the implicit MPFconverges faster and is more accurate than a conventional particle filterduring periods of persistent contact. We present three methods of sampling theimplicit contact manifold, and compare them in experiments.
展开▼
