Safety in dynamical systems is typically binary: safe or violated. A control barrier function keeps the system inside a safe set. If it fails, the system is outside the set. There's no middle ground.
Note (2026-11-07): the abstract builds a safety constraint combining zeroing and reciprocal control barrier functions for a primary and a secondary safe set, analogous to a stiffening spring, for systems of relative degree 1 or 2, and demonstrates it on one wall collision avoidance example against traditional benchmarks. The essay's explanation of why it outperforms (not spending control effort defending the primary boundary) and its general claim about any domain where perfect standards are costly are my own.
This paper introduces a middle ground by layering two barrier functions — a primary layer representing the desirable safety standard and a secondary failsafe layer. The primary can be violated. The secondary cannot. Between them lies a region of degraded but survivable operation.
The energy-based framework makes this concrete: zeroing barrier functions handle the primary layer, reciprocal barrier functions handle the failsafe. The combination ensures that if the system leaves its preferred operating region, it enters a controlled degradation zone rather than falling off a cliff.
Applied to wall collision avoidance, the approach outperforms existing methods because it doesn't waste control effort defending the primary boundary at all costs. It can let the primary boundary go, gracefully, knowing the failsafe catches it.
The structural insight is about the topology of safety. Binary safety creates a single boundary that must be defended everywhere. Layered safety creates a buffer zone that absorbs violations without catastrophe. The system that allows partial failure performs better than the system that treats all failure as equally unacceptable.
This maps onto any domain where the cost of maintaining perfect standards exceeds the cost of occasional graceful degradation.