All robots differ from their idealized mathematical models because of parametric errors, link flexion, environmentally induced errors, etc. Because the resulting equations are too complex, analytically derived models cannot include compensation for such errors. Numerical models, however, can easily provide compensation.During Phase I, the feasibility of providing compensation in a quasi-static situation was demonstrated by deriving multi-dimensional, B-spline models of inverse kinematic equations for a robot. These models were shown to be immune to certain parametric errors, measurement noise, and the dimensionality of the functions. The partial derivatives of the models also seem accurately to represent the true joint partial derivatives. Some errors (payload induced link flexion, axes misalignment, etc.) are too complex to simulate. However, if models are derived from measured end effector and joint data, which include the effects of such errors, error compensation will implicitly be included in the models.Potential Commercial Application:Robot models that include error compensation can be substituted for existing models in operational robots and are particularly appropriate for controllers marketed for use with various robots. The modeling algorithm will also make it possible to use flexible or imprecisely manufactured robots for precision tasks.STATUS: Phase I Only
