[comp.ai.digest] Seminar - Uncertainties in Robot Planning

jutta@WHITNEY.STANFORD.EDU (Jutta McCormick) (07/30/87)

Thursday, August 6, 2:30 p.m.
Robotics Lab Conference Room, Cedar Hall, A6.


           DEALING WITH UNCERTAINTIES IN ROBOT PLANNING
                USING PROGRAM PROVING TECHNIQUES

                  Dr. Jocelyne Pertin-Troccaz
                    LIFIA Laboratory - INPG
                    46, Avenue Felix Viallet
                 38031. Grenoble Cedex - France
              (uucp: lifia!jocelyne@seismo.css.gov)


Abstract: A global approach for dealing with uncertainties using
program proving in Robotics is presented. We consider a manipulation
program automatically generated by a planner according to spatial and
geometric criteria and ignoring uncertainties. Such a program is
correct only if, at each step, uncertainties are smaller than the
tolerances imposed by the assembly task. We propose an approach which
consists in verifying the correctness of the program with respect to
uncertainties in position and possibly modifying it by adding
operations in order to reduce uncertainties. These two steps based on
a forward and a backward propagation borrowed from formal program
proving techniques are described in a general framework suitable for
robotics environments. Forward propagation consists in computing
successive states of the robot world from the initial state and in
checking for the satisfaction of constraints. If a constraint is not
satisfied, backward propagation infers new constraints on previous
states. These new constraints are used for patching the program. The
approach is described in technical details in the case of a simple
manipulation language and of a relational model of the world including
a representation of uncertainties.


Thanks.
-Jutta
---- GMT
S GMT
S