Introduction
In modern robotics, ensuring safety is paramount. Control Barrier Functions (CBFs) are a standard mathematical framework used to guarantee that a system remains within a designated "safe set." Typically, these are implemented through a Quadratic Program (QP) that calculates the optimal control command while satisfying safety constraints. However, real-world robots face physical limitations, such as motor saturation or maximum braking force. A critical failure occurs when these physical limits conflict with safety requirements, leading to point-wise infeasibility. In such cases, the optimization problem has no solution, and the controller fails to provide any command, potentially leading to catastrophic system failure.
The Optimal-decay CBF Formulation
To overcome the challenge of infeasibility, this research introduces the Optimal-decay Control Barrier Function. In a standard CBF setup, safety is enforced using a fixed decay rate, $\alpha$, which dictates how quickly the system must move away from the boundaries of the safe set. If the required maneuver exceeds the robot's physical capabilities, the system becomes infeasible.
The proposed solution introduces a new decision variable, $\omega$, which acts as a scaling factor for the decay rate. The safety constraint is reformulated as: $$L_f h(x) + L_g h(x)u \geq -\omega \alpha(h(x))$$ By incorporating $\omega$ into the QP's objective function with a penalty term, the optimizer can dynamically adjust the "stiffness" of the safety constraint. This provides a "pressure release valve": when physical limits are reached, the system can increase $\omega$, effectively relaxing the safety margin just enough to find a valid, executable control command. This approach guarantees point-wise feasibility for any state strictly within the safe set.
Geometric Interpretations of Feasibility
The paper elucidates the problem of infeasibility through two distinct geometric lenses: the input space and the state space.
Input-Space Perspective
Feasibility in the input space is determined by the intersection of the set of admissible controls $\mathcal{U}_{adm}$ (the robot's physical limits) and the set of safe controls $\mathcal{U}_{cbf}$ (the commands that satisfy the safety constraint).
As illustrated in the figure above, infeasibility occurs when the safety requirement (the red line) shifts such that it no longer overlaps with the polygon of admissible controls. The optimal-decay method allows the safety boundary to move, ensuring a non-empty intersection.
State-Space Perspective
In the state space, feasibility depends on whether the system's reachable set—the collection of all states it can reach in the next time step—overlaps with the safe region defined by the CBF.
The sequence above demonstrates how the reachable set $\mathcal{R}$ (yellow) interacts with the safe set boundary. As the system approaches a state where the reachable set lies entirely outside the safe boundary (panel d), a standard controller would fail. The optimal-decay strategy prevents this by expanding the allowed reachable set through the adjustment of $\omega$.
Validation: Adaptive Cruise Control (ACC)
The authors validated their approach using an Adaptive Cruise Control simulation. The "ego" vehicle was tasked with maintaining a safe distance from a lead vehicle while respecting strict acceleration and braking limits.
In high-speed scenarios (e.g., 32 m/s), the standard CBF controller became infeasible because the required braking to maintain safety exceeded the vehicle's physical limits, causing the simulation to halt. In contrast, the Optimal-decay controller successfully navigated the same scenario. By allowing the variable $\omega$ to spike momentarily, the controller utilized maximum braking power without the optimizer failing. When the vehicle returned to a lower-stress state, $\omega$ returned to its nominal value of 1, and the controller behaved like a standard CBF.
Conclusion
The Optimal-decay CBF-QP offers a robust framework for safety-critical control by guaranteeing point-wise feasibility. By making the safety decay rate an optimization variable, it ensures that autonomous systems can always find a valid control command, even under extreme physical constraints. While this method addresses immediate feasibility, future work will focus on "persistent feasibility" to ensure that current safe actions do not lead to unavoidable future collisions.
Summary updated