Constraint relaxation to reduce brittleness of distributed agent protocols

Incompatible goals among multiple agents
working on domains involving finite constraints can be a
source of conflict. This conflict, in the form of
incompatible constraints established locally by the agents
and imposed on the negotiated variables, may break the
dialogue between these agents even though they could, in
principle, reach an agreement. A common means of
coordinating multi-agent systems is by using protocols to
which are attached constraints on interaction; but
protocols are brittle, in the sense that the constraints they
contain must either succeed or fail, and if they fail the
entire protocol may fail. We apply a constraint relaxation
technique originally for automated negotiation using a
distributed protocol language called the Lightweight
Coordination Calculus (LCC), in order to overcome a
class of conflicts, making protocols less brittle. This
approach is illustrated in a scenario involving the
ordering and configuration of a computer between the
customer and vendor agents.