Control Barrier Function (CBF)

Forward invariance is the foundational guarantee provided by a valid CBF. If a system starts inside a safe set defined by the CBF, and the controller satisfies the CBF condition for all future time, then the system is guaranteed to remain inside the safe set forever. This property transforms a static safety specification into a dynamic control law.

• Mathematical Definition: A set (C) is forward invariant if for every initial state (x(0) \in C), the trajectory (x(t) \in C) for all (t \geq 0).
• Role of the CBF: The CBF's time derivative condition, (\dot{h}(x) \geq -\alpha(h(x))), is a constraint that, when enforced, mathematically proves forward invariance.

The relative degree of a CBF is the number of times the output function (h(x)) must be differentiated with respect to time before the control input (u) explicitly appears. This property dictates the form of the CBF constraint and the complexity of the resulting safety filter.

• Relative Degree 1: The control input (u) appears in the first derivative (\dot{h}(x,u)). This leads to a simple, affine constraint in (u) that is easy to enforce via a Quadratic Program (QP). Example: Velocity-level constraints for a mobile robot.
• High Relative Degree: Requires (u) to appear after multiple differentiations. This necessitates techniques like Exponential Control Barrier Functions (ECBFs) or Integral Control Barrier Functions to formulate a constraint that is still applicable for real-time control.

A primary use of a CBF is as a safety filter. It takes a potentially unsafe nominal control input (e.g., from an MPC or tracking controller) and minimally modifies it to ensure safety. This is typically formulated as a Quadratic Program (QP).

CBF-QP Formulation: [\min_{u} , |u - u_{nom}|^2] [\text{subject to: } , L_f h(x) + L_g h(x) u \geq -\alpha(h(x))]

• (u_{nom}): The desired, performance-optimizing input from the nominal controller.
• CBF Constraint: The inequality that enforces the forward invariance condition.
• Minimal Intervention: The QP finds the safest control input closest to the nominal one, preserving performance when safety is not at risk.

The class (\mathcal{K}) function, (\alpha(\cdot)), in the CBF condition (\dot{h} \geq -\alpha(h)) is a tunable parameter that dictates how aggressively the controller acts to keep the system safe.

• Function Definition: A continuous function (\alpha: [0, \infty) \to [0, \infty)) is class (\mathcal{K}) if it is strictly increasing and (\alpha(0) = 0). A common choice is (\alpha(h) = \gamma h) for (\gamma > 0).
• Interpretation: When the system is far from the boundary ((h) is large), the constraint (\dot{h} \geq -\gamma h) is loose, allowing the nominal controller to act freely. As the system approaches the boundary ((h \to 0)), the constraint forces (\dot{h}) to become positive, pushing the state back into the interior of the safe set. A larger (\gamma) results in more aggressive corrective action.

CBFs and MPC can be integrated in two primary architectures, leveraging the predictive nature of MPC with the formal safety guarantees of CBFs.

1. CBFs as Constraints in MPC: The CBF condition (\dot{h}(x,u) \geq -\alpha(h(x))) is added as a constraint within the MPC's finite-horizon Optimal Control Problem (OCP). This ensures the optimized trajectory is safe over the prediction horizon.
2. CBFs as a Safety Filter for MPC: The MPC solves its OCP without explicit safety constraints to maximize performance. The resulting planned input sequence is then passed through a CBF-based safety filter (e.g., a CBF-QP) at each time step before being applied to the system. This decouples performance optimization from safety enforcement.

CBFs are often discussed alongside Control Lyapunov Functions (CLFs), which certify stability (convergence to a goal). Understanding their dual role is key to unified controller design.

• CLF-CBF QP: A single Quadratic Program can enforce both CLF (for performance) and CBF (for safety) constraints simultaneously, providing a unified framework for safe and stable control.

A Lyapunov function is a scalar, energy-like function used to prove the asymptotic stability of an equilibrium point for a dynamical system. While a CBF certifies safety (state remains within a set), a Lyapunov function certifies stability (state converges to a point). They are often used together in Control Lyapunov-Barrier Functions (CLBF) to achieve both stability and safety guarantees.

• Key Property: Must be positive definite and its derivative along system trajectories must be negative definite.
• Analogy: If a CBF defines a "safe basin," a Lyapunov function defines the "bottom of the bowl" the system settles into.

A Control Lyapunov Function (CLF) is a Lyapunov function whose time derivative can be made negative by an appropriate choice of control input. It is a constructive tool for stabilization and tracking. In combined frameworks like CLF-CBF Quadratic Programs (QPs), CLFs and CBFs are used as competing constraints to find a control input that simultaneously stabilizes the system while respecting safety barriers.

• Primary Use: Guaranteeing convergence to a desired state or trajectory.
• Formulation: Requires finding a control u such that the derivative of the CLF is negative.

An invariant set is a region in the state space such that if the system state starts inside it, it remains inside for all future time under a given control law. The core function of a CBF is to render a safe set (defined by the CBF's superlevel set) forward invariant.

• Forward Invariance: The property guaranteed by a valid CBF.
• Types: Includes positively invariant sets, control invariant sets, and robust control invariant sets.
• Design Goal: The safe set {x | h(x) ≥ 0} is designed to be a control invariant set.

A barrier certificate is a related formal verification concept for continuous dynamical systems. It is a scalar function whose zero-level set forms a barrier between an initial set and an unsafe set, proving that no trajectory starting in the initial set can reach the unsafe set. While similar in spirit to a CBF, barrier certificates are typically used for offline analysis of a fixed closed-loop system, whereas CBFs are used for online synthesis of a safe controller.

• Verification vs. Synthesis: Barrier certificates verify; CBFs synthesize.
• Application: Proving safety of existing autonomous systems or neural network controllers.

Hamilton-Jacobi (HJ) Reachability is a formal method for computing the Backward Reachable Tube (BRT)—the set of all states from which the system can be driven into an unsafe set within a time horizon, despite the best control efforts. It provides the ultimate safety guarantee but is computationally intensive for high-dimensional systems. CBFs can be seen as a simplifying, controller-synthesizing alternative; the safe set for a CBF is often designed as a subset of the complement of the BRT.

• Exact vs. Conservative: HJ provides exact reachable sets; CBFs provide efficient, sometimes conservative, safety filters.
• Curse of Dimensionality: HJ is limited to ~5-6 state dimensions, while CBFs scale better.

An Exponential Control Barrier Function (ECBF) is a higher-order extension of the standard CBF used for systems with relative degree greater than one. It enforces safety by ensuring not just that h(x) ≥ 0, but that a set of its time derivatives also satisfy specific conditions, effectively shaping the approach to the barrier boundary. This is critical for physical systems where the control input (e.g., acceleration) does not directly affect the safety constraint (e.g., position).

• Relative Degree: The number of times h(x) must be differentiated before the control input u appears explicitly.
• Application: Essential for robotic systems where force/torque controls position constraints.