Reciprocal Resistance-Based Control Barrier Function

• RRCBF is a control barrier framework that integrates a reciprocal resistance term to robustly enforce forward invariance of interior safe sets.
• It creates a buffer zone by augmenting classical barrier inequalities, steering system trajectories away from unsafe boundaries.
• The formulation extends to high-order and disturbance-observer variants, enabling effective QP-based safety filtering without explicit disturbance bounds.

Searching arXiv for the cited RRCBF paper and closely related CBF background papers. {"query":"arXiv (Wang et al., 25 Jul 2025) Reciprocal Resistance-based control barrier function (Spiller et al., 24 Mar 2025, Ames et al., 2016)", "max_results": 10} Using the arXiv search tool to retrieve the relevant records. arXiv search query: "(Wang et al., 25 Jul 2025)" Reciprocal Resistance-Based Control Barrier Function (RRCBF) denotes a control-barrier construction for disturbed affine nonlinear systems in which the conventional zeroing-barrier inequality is augmented by a reciprocal resistance-like term of the form $\beta(1/h(x))$. In the formulation introduced in "Enhancing Robustness of Control Barrier Function: A Reciprocal Resistance-based Approach" (Wang et al., 25 Jul 2025), this term grows unboundedly as the safety function $h(x)\to 0^+$, thereby creating a buffer zone near the boundary of the safe set and enabling robustness against bounded disturbances without requiring explicit prior knowledge of disturbance bounds in the controller design. The framework includes the underlying reciprocal resistance-based barrier function (RRBF), its control version RRCBF, high-order extensions for relative-degree-$r$ constraints, and a disturbance observer-based variant.

1. Terminology and conceptual lineage

The term RRCBF is recent and should be distinguished from earlier barrier-function nomenclature. In "Control Barrier Function Based Quadratic Programs for Safety Critical Systems" (Ames et al., 2016), reciprocal control barrier function (RCBF) refers to a reciprocal barrier $B(x)$, typically $B(x)=1/h(x)$ or $B(x)=-\log(h(x)/(1+h(x)))$, together with an inequality on $L_f B + L_g B\,u$. That framework does not use the term resistance-based.

A second source of possible ambiguity arises in "Feasibility of multiple robust control barrier functions for bounding box constraints" (Spiller et al., 24 Mar 2025). There, RCBF means robust control barrier function, and the relevant construction is a resistance-based augmented barrier

$$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$

which incorporates a braking-distance term. That paper explicitly does not analyze a reciprocal transformation $b=1/H$.

Against that background, the 2025 note (Wang et al., 25 Jul 2025) introduces a distinct object: the reciprocal resistance-based barrier function (RRBF) and its control counterpart RRCBF. Its characteristic feature is not a reciprocal change of variables alone, but the insertion of a reciprocal resistance-like term directly into the barrier inequality,

$$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$

for extended class-$h(x)\to 0^+$0 functions $h(x)\to 0^+$1 and $h(x)\to 0^+$2. This places RRCBF at the intersection of two earlier strands: reciprocal-barrier sensitivity near the boundary and robust barrier shaping against disturbances.

2. Core formulation and buffer-zone mechanism

The formulation in (Wang et al., 25 Jul 2025) considers two disturbance models. For matched disturbances, the control design is based on

$h(x)\to 0^+$3

where $h(x)\to 0^+$4, $h(x)\to 0^+$5, $h(x)\to 0^+$6 and $h(x)\to 0^+$7 are locally Lipschitz on the compact admissible state set $h(x)\to 0^+$8, and $h(x)\to 0^+$9 for some unknown constant $r$0. For unmatched disturbances, used to establish RRBF robustness at the barrier level, the model is

$r$1

Safety is encoded by a continuously differentiable function $r$2, with

$r$3

For comparison, the conventional zeroing CBF condition for relative degree one is

$r$4

for some extended class-$r$5 function $r$6.

The motivating scalar mechanism is the disturbed differential equation

$r$7

with $r$8, $r$9, and $B(x)$0. The reciprocal term $B(x)$1 diverges as $B(x)$2 and prevents loss of positivity. Under the worst-case disturbance $B(x)$3, one obtains $B(x)$4 for $B(x)$5, where

$B(x)$6

Under $B(x)$7, one has $B(x)$8 for $B(x)$9, with

$B(x)=1/h(x)$0

This provides the paper’s template for disturbance domination near the boundary.

For the autonomous system $B(x)=1/h(x)$1, an RRBF is a continuously differentiable $B(x)=1/h(x)$2 such that there exist extended class-$B(x)=1/h(x)$3 functions $B(x)=1/h(x)$4 and $B(x)=1/h(x)$5 satisfying

$B(x)=1/h(x)$6

Equivalently,

$B(x)=1/h(x)$7

For the disturbed affine system, the corresponding RRCBF condition is

$B(x)=1/h(x)$8

with admissible-control set

$B(x)=1/h(x)$9

A central geometric notion is the buffer zone. Let $B(x)=-\log(h(x)/(1+h(x)))$0 be the unique solution of

$B(x)=-\log(h(x)/(1+h(x)))$1

Then define

$B(x)=-\log(h(x)/(1+h(x)))$2

The region

$B(x)=-\log(h(x)/(1+h(x)))$3

is the buffer zone in which the reciprocal term dominates and drives trajectories toward $B(x)=-\log(h(x)/(1+h(x)))$4 rather than toward $B(x)=-\log(h(x)/(1+h(x)))$5.

3. Forward invariance and disturbance robustness

The fundamental invariance statement in (Wang et al., 25 Jul 2025) is that if $B(x)=-\log(h(x)/(1+h(x)))$6 is an RRBF for $B(x)=-\log(h(x)/(1+h(x)))$7 and $B(x)=-\log(h(x)/(1+h(x)))$8, then $B(x)=-\log(h(x)/(1+h(x)))$9 is forward invariant. The proof splits the state space into the robust subset $L_f B + L_g B\,u$0 and the buffer zone. On $L_f B + L_g B\,u$1, the RRBF inequality gives $L_f B + L_g B\,u$2, so $L_f B + L_g B\,u$3 is forward invariant by Nagumo’s theorem. For initial conditions in $L_f B + L_g B\,u$4, the function

$L_f B + L_g B\,u$5

satisfies

$L_f B + L_g B\,u$6

which implies convergence toward $L_f B + L_g B\,u$7.

The robustness theorem extends this argument to unmatched bounded disturbances. If

$L_f B + L_g B\,u$8

then

$L_f B + L_g B\,u$9

Because $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$0 is bounded on compact $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$1 and $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$2 is bounded, there exists $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$3 such that $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$4, hence

$$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$5

If $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$6 is the unique solution of

$$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$7

then $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$8 whenever $$H(t,x)=h(t,x)+\frac{|\dot h(t,x)|\dot h(t,x)}{2a_{\max}},$$9. This excludes exit through the boundary and preserves positivity of $b=1/H$0.

For matched disturbances, if $b=1/H$1, then

$b=1/H$2

and the same dominance argument applies by boundedness of $b=1/H$3 and $b=1/H$4. An important point in the paper is that controller synthesis does not require explicit knowledge of $b=1/H$5 or $b=1/H$6; the guarantee depends only on the existence of finite disturbance bounds and on the divergence of $b=1/H$7 as $b=1/H$8.

4. High-order and disturbance-observer extensions

For safety constraints of relative degree $b=1/H$9, (Wang et al., 25 Jul 2025) defines a high-order RRCBF (HO-RRCBF). Let $$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$0 satisfy

$$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$1

Define recursively

$$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$2

and the sets

$$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$3

with

$$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$4

Using

$$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$5

the HO-RRCBF condition is

$$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$6

for all $$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$7.

The resulting invariance theorem states that if a Lipschitz controller satisfies the HO-RRCBF constraint and the initial conditions obey $$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$8 for $$L_f h(x)+L_g h(x)\,u+\alpha(h(x))-\beta\!\left(\frac{1}{h(x)}\right)\ge 0,$$9, then $h(x)\to 0^+$00 is forward invariant. The proof reduces the highest-order condition to an RRBF-type inequality in the variable $h(x)\to 0^+$01 and then propagates invariance down the recursion.

The same paper introduces a disturbance observer-based RRCBF (DO-RRCBF). With disturbance estimate $h(x)\to 0^+$02 and estimation error $h(x)\to 0^+$03, a typical local error model is

$h(x)\to 0^+$04

with suitable gain $h(x)\to 0^+$05. The DO-RRCBF inequality becomes

$h(x)\to 0^+$06

and the corresponding admissible set is

$h(x)\to 0^+$07

A stated advantage is that DO-RRCBF does not require an explicit bound on the estimation error in the barrier constraint. The reciprocal resistance term is used to dominate residual estimation error near the boundary while the disturbance estimate reduces conservatism away from it.

5. Controller synthesis, tuning, and limitations

The implementation in (Wang et al., 25 Jul 2025) is QP-based. A CLF-CBF formulation combines stabilization via a control Lyapunov function $h(x)\to 0^+$08 and safety via the RRCBF constraint:

$h(x)\to 0^+$09

Here $h(x)\to 0^+$10 is a slack variable for the CLF constraint and $h(x)\to 0^+$11 is its weight.

If a nominal controller $h(x)\to 0^+$12 is already available, the barrier can be used as a safety filter:

$h(x)\to 0^+$13

The high-order variant is imposed analogously by replacing the relative-degree-one inequality with the HO-RRCBF constraint.

The paper repeatedly emphasizes linear choices

$h(x)\to 0^+$14

Larger $h(x)\to 0^+$15 accelerates convergence in the interior but can increase control effort. Larger $h(x)\to 0^+$16 strengthens disturbance dominance near $h(x)\to 0^+$17 but increases conservatism and numerical stiffness. The buffer threshold $h(x)\to 0^+$18 is determined by $h(x)\to 0^+$19; according to the paper, reducing $h(x)\to 0^+$20 increases $h(x)\to 0^+$21 and yields a less conservative, larger robust subset $h(x)\to 0^+$22.

The principal numerical difficulty is the singularity as $h(x)\to 0^+$23. The proposed regularization is to replace $h(x)\to 0^+$24 by $h(x)\to 0^+$25 with small $h(x)\to 0^+$26, leading to

$h(x)\to 0^+$27

Formal guarantees with this regularization require a disturbance bound $h(x)\to 0^+$28 on $h(x)\to 0^+$29 and a choice

$h(x)\to 0^+$30

If $h(x)\to 0^+$31 is unknown, the paper suggests choosing a small $h(x)\to 0^+$32 and validating empirically.

The limitations identified in (Wang et al., 25 Jul 2025) are standard but consequential: singularity and numerical stiffness near the boundary, feasibility degradation under tight control limits, the fact that the RRCBF definition itself presumes matched disturbances in the control channel, and sensitivity of DO-RRCBF performance to observer tuning.

6. Relation to adjacent CBF frameworks and reported case studies

Relative to ZBF/ZCBF, the RRCBF construction preserves the zeroing-barrier structure in $h(x)\to 0^+$33 but adds the reciprocal resistance term $h(x)\to 0^+$34. The paper’s stated comparison is that ZBFs enforce invariance of $h(x)\to 0^+$35 but may be sensitive near $h(x)\to 0^+$36, whereas RRBF/RRCBF enforce forward invariance of $h(x)\to 0^+$37 and create a buffer zone in which the reciprocal term dominates disturbances. Relative to classical reciprocal CBFs in the sense of $h(x)\to 0^+$38 (Ames et al., 2016), RRCBF retains reciprocal sensitivity near the boundary but introduces an explicit disturbance-dominating term rather than only an inequality on $h(x)\to 0^+$39. This suggests a different robustness mechanism: not merely barrier blow-up, but direct reciprocal forcing in the $h(x)\to 0^+$40-dynamics.

The distinction from the resistance-based robust CBF for bounding-box constraints in (Spiller et al., 24 Mar 2025) is structural. That work augments $h(x)\to 0^+$41 with a braking-distance term $h(x)\to 0^+$42 to form $h(x)\to 0^+$43 and derives feasibility conditions for multiple robust half-space constraints under input bounds and disturbances. By contrast, (Wang et al., 25 Jul 2025) augments the barrier inequality itself with $h(x)\to 0^+$44 and does not rely on the $h(x)\to 0^+$45-type braking-distance construction. The two uses of “resistance-based” therefore refer to different mechanisms.

The stochastic literature provides yet another nearby concept. "Control Barrier Functions for Stochastic Systems and Safety-critical Control Designs" (Nishimura et al., 2022) studies reciprocal and zeroing control barrier functions for stochastic systems, with reciprocal barrier $h(x)\to 0^+$46 and generator inequality $h(x)\to 0^+$47 for almost-sure-type safety in the paper’s FIiP sense. That framework is reciprocal but not resistance-based in the specific sense of (Wang et al., 25 Jul 2025).

The simulations reported in (Wang et al., 25 Jul 2025) are consistent with these distinctions. For the second-order linear system

$h(x)\to 0^+$48

with nominal controller $h(x)\to 0^+$49, $h(x)\to 0^+$50, $h(x)\to 0^+$51, and disturbance $h(x)\to 0^+$52, the paper compares ZCBF, RCBF with $h(x)\to 0^+$53, and RRCBF. It reports that without disturbance, RRCBF renders $h(x)\to 0^+$54 invariant, with trajectories starting in $h(x)\to 0^+$55 remaining there and those starting in the buffer zone converging to $h(x)\to 0^+$56. With disturbance, ZCBF and RCBF allow trajectories to exit the safe set, whereas RRCBF prevents boundary violation.

The adaptive cruise control study addresses a relative-degree-$h(x)\to 0^+$57 safety constraint:

$h(x)\to 0^+$58

with safety function $h(x)\to 0^+$59, desired speed $h(x)\to 0^+$60, $h(x)\to 0^+$61, initial $h(x)\to 0^+$62, $h(x)\to 0^+$63, disturbance $h(x)\to 0^+$64, control bound $h(x)\to 0^+$65, and gains $h(x)\to 0^+$66, $h(x)\to 0^+$67, $h(x)\to 0^+$68, observer gain $h(x)\to 0^+$69. The reported outcome is that RRCBF, DO-RRCBF, and RCBF with known disturbance bounds maintain strict safety under disturbances; DO-RRCBF reduces conservatism by compensating estimated disturbances and recovering nominal performance; and DO-CBF without an explicit estimation-error bound fails to maintain strict safety.

Taken together, these results position RRCBF as a specific 2025 development within the broader CBF literature: a reciprocal-resistance augmentation of zeroing-barrier inequalities designed to preserve forward invariance of the interior safe set under bounded disturbances, to extend naturally to higher relative degree, and to admit standard QP-based safety filtering and CLF-CBF synthesis (Wang et al., 25 Jul 2025).