Finer-grained interruption of atomic end-to-end invocations

Develop finer-grained interruption mechanisms for atomic end-to-end policy invocations under ambiguous observations, so that responsibility can change before an invocation reaches a postcondition, reports failure, or times out, rather than only at predefined semantic stage boundaries.

Background

RoboAware coordinates modular composition and end-to-end policies by selecting a policy family at semantic boundaries defined by the P^5 Skill Schema. Once an atomic end-to-end call begins, the selected family retains control across potentially several stages until a postcondition, failure signal, or timeout returns control to the coordinator.

The authors identify this granularity as a limitation: transient failures occurring within an atomic end-to-end invocation cannot trigger an earlier handoff when observations are ambiguous. They explicitly leave the development of finer-grained interruption as unresolved future work.