A fuzzy logic approach to contact identification M. Oussalah, H. Bruyninckx, J. De Schutter K.U of Leuven, Department of Mechanical Engineering Division PMA, Celestjnenlaan 300B, 3001 Heverlee, Belguim Tel: +32 16 32 25 28 Fax: +32 16 32 29 87 Abstract This paper discusses some compliant motion tasks used for geometry identification pertaining to some contact situations. Particularly, we focus on the simple vertex-face situation where the input parameters are force, incremental moving and velocity, each with six components (three translations and three rotations). This analysis is based on fuzzy logic approach where the data suitably modelled in terms of possibility distributions and the results are performed like fuzzy numbers computation in the sprit of Kalman filter approach. The fuzzy computation is approximated such that L-R fuzzy number representations are performed iteratively. Moreover, the approach supplies an interval-like method where upper and lower bound of Kalman state and covariance estimate are efficiently computed. IMACS/IEEE CSCC'99 Proceedings, Pages:5791-5797 1-Introduction In compliant or constrained robot motion, the robot moves the manipulated object while it is in contact with other objects in the environment. In force controlled compliant motion, most research focusses on the lowest servo control level (how to control simultaneously the motion of the manipulated object and the contact forces), and much less on “intelligent” sensing (to determine the geometry and/or type of the contact situation). Vertex-fac e is an example of a type of contact situation; the positio n of the contact point, and the orientatio n of the contact normal are examples of geometric parameters in the given contact type. Previous work by the authors, [1-4], presented a general theory to describe all possible contact situations, to specify any desired compliant motion task in terms of desired contact forces and desired velocity setpoints, and to find a set of equations linking the measured force and velocity with the geometric uncertainty parameters in the contact model. This parameter estimation uses Extended Kalman Filters. This paper treats the same problem using a fuzzy approach, (see [5] for more discussion about the theory). In particular, the Kalman filter approach is imitated in its “formulation aspect”: the parameters to be estimated are represented in terms of possibility distributions, and are computed using an L-R representation [6], which leads to an iterative process. The result links appropriately with interval analysis, and provides an appealing interpretation. In order to focus on the main characteristics of the approach, a simple case study is considered, namely a vertex-face contact. Experimental results are presented. This paper is organised as follows. Section 2 introduces the Kalman filter approach, and applies it to the vertex-face contact. Next possibility theory with required background is introduced; the manipulation of possibilistic quantities by means of LR representation and given approximation is also introduced. Then, the method for handling the contact situation problem is described and their results discussed. 2- Kalman filter approach A Kalman filter [7] is a recursive optimal estimator which minimizes the variance of the estimate under a linearity assumption, and when both state and measurements are corrupted by zero mean gaussian noise. However, when the system is non-linear, the Kalman approach can be extended by linearizing around the state estimate, i.e., using the Jacobien of the non-linear system.