Skip to Main Content
The need to equip robotic systems with a high degree of dexterity and adaptability to a wide range of precise manipulation tusks has given rise to increasing interest in the utilization of various sensing modalities, including force and tactile sensing. In this paper a general framework is proposed for incorporating both a priori task geometry information and on-line observations, including force measurements, into an optimal estimation algorithm. The output of the algorithm is state and parameter estimates that serve to disambiguate the task geometry and can be used to dynamically adapt subsequent motions. The problem is formulated us a nonlinear constrained dynamical system, including Coulomb friction between the system and the constraints. The constraint surface is described with respect to some unknown parameters representing the geometric uncertainty. The noisy on-line state and force observations are expressed as functions of the state and surface parameters. The extended Kalman filter is then used to produce optimal estimates of the state and surface parameters.