A Geometric State Sensor for Robot Workcell Monitoring

Abstract

A processing scheme is given that provides data at a point along a sensory continuum which is distinct from the region normally occupied by those data provided by the sensing devices more conventionally used in robotic assembly. In contrast to the operation of typical binary sensors such as contact probes and beam-interruption devices, which give an unambiguous datum about a local event in the cell, the sensor described is sensitive to deviations of any component in the workcell volume with an implementation dependent resolution, presenting no other information except that a deviation has occurred.

Where required in. the assembly sequence, a signature is derived by a simple off-line teaching procedure. Although the use of the sensor in an analytic role is not expected to be tenable, it provides a cue for the onset of error diagnosis in a cell supervisory computer, operating either in stand-alone mode or integrated with other sensory input.

An argument from the field of statistical communication theory is presented, showing the eliciration of system behaviour from a communication theoretic formalism to be inappropriate for robot domains due to the nature of the knowledge representation. The use of domain specific knowledge is discussed with a view to establishing likely bounds for probabilistic descriptors of the scene.

The A.I technique of ‘evidential propositional calculus’ is applied for the combination of multiple independent sensor readings.