Readiness Barrier Functions: Forward-Invariant Control Authority for Overactuated Multirotor Allocation

What the math gives to ML

The paper turns redundancy into a certified dynamical resource: a log-determinant authority metric is kept above a floor by a forward-invariance barrier rather than optimized greedily at each step. Its transferable asset is the combination of a diversity-sensitive log-det potential, an exact rank/leverage interpretation of component dropout, and a null-space quadratic program that changes an allocation as little as possible while enforcing the barrier. A promising neural analogue is an MoE router whose expert contribution directions remain sufficiently diverse, preventing routing collapse while preserving the requested mixture output. The method is directly testable as a constrained router update and provides interpretable leverage scores identifying experts whose removal would most damage representational authority.