Imagine driving a vehicle toward an obstacle. To guarantee safety, you must ensure the vehicle never crosses a spatial boundary—meaning its position must remain within a designated safe set. In a foundational overview of safety-critical control, Ames et al. describe how standard Control Barrier Functions (CBFs) achieve this by ensuring that the first derivative of the safety metric can always be directly manipulated by the control input. If the vehicle gets too close to the boundary, the controller instantly dictates a safe velocity to steer it away.
However, physical systems do not work this way. You cannot instantaneously change a vehicle's velocity; you can only change its acceleration by applying torque to the brakes or motor (Kaypak et al.). Because the control input (acceleration) does not appear in the first derivative of the safety constraint (velocity), a standard CBF breaks down. This mismatch is known as a high relative degree problem. If a safety filter only reacts based on current position, the vehicle's momentum will carry it through the obstacle before the brakes can take effect.
To resolve this lag between actuation and safety outcomes, researchers developed High-Order Control Barrier Functions (HOCBFs).
The Core Mechanism: Differentiating to Reveal Control
The fundamental insight behind HOCBFs is that if a control input does not immediately influence a safety constraint, we must differentiate the constraint multiple times until the control input explicitly appears (Wang et al.). For a system where the control input is revealed at the $m$-th derivative, the safety constraint is said to have a relative degree of $m$.
By calculating successive Lie derivatives of the safety function (Al Maruf et al.), the framework constructs a chain or sequence of nested safety functions (Zinage et al.). Each function in this sequence defines a safe operating region for the derivative of the preceding function.
For our vehicle example, this means safety margins are distributed across multiple system states simultaneously (Zhang et al.):
- Position: The vehicle must stay a certain distance from the wall.
- Velocity: As the vehicle approaches the wall, its allowable forward velocity must decrease.
- Acceleration: If the vehicle is moving quickly toward the wall, the controller must enforce a strict deceleration command.
By linking these states together, the higher-order derivatives act as a predictive buffer, forcing the system to take preemptive action long before it reaches the physical boundary.
Ensuring Forward Invariance
A set is considered forward invariant if a system starting inside the set is mathematically guaranteed to never leave it. As Malikopoulos demonstrates, HOCBFs enforce forward invariance for high-relative-degree systems by constraining the highest-order derivative through linear feedback techniques.
To implement this without the burden of heavy real-time optimization, Gao utilizes Exponential Control Barrier Functions (ECBFs). By choosing a sequence of tuning parameters (often structured via class $\alpha$ functions), the controller forces the higher-order derivatives to behave like a stable linear system. As the state approaches the unsafe boundary, the higher-order constraints tighten exponentially, smoothly overriding the user's inputs to guarantee safety without causing jerky or discontinuous control actions (Almubarak et al.).
Managing Complexities and Multi-Input Systems
Real-world physical systems frequently introduce asymmetric constraints where different control inputs influence safety at different relative degrees. For example, in autonomous driving, a steering command (lateral control) might affect the distance to an obstacle faster or slower than a braking command (longitudinal control).
To prevent the system from being overly restricted to a single control dimension, Xiao et al. introduce auxiliary dynamics. By treating the actual control inputs as internal system states and using their derivatives as virtual controls, they artificially increase the differentiation order. This mathematical adjustment ensures that every available control input is represented within the safety constraint formula, allowing the system to utilize all degrees of freedom—such as steering around an obstacle when braking alone is insufficient.
When the underlying physics of the system are complex or imperfectly understood, calculating exact high-order derivatives becomes difficult. To handle this, Wu et al. and Zhang et al. combine HOCBFs with Gaussian Process (GP) regression. The GP learns model mismatches and external disturbances in real time, allowing the HOCBF to adaptively tighten or loosen its safety boundaries based on the uncertainty of the environment.
Alternatively, to bypass the tedious symbolic math of calculating high-order Lie derivatives altogether, Almubarak et al. propose embedding constraints directly into the system dynamics as "barrier states." This transformation effectively reformulates a high-relative-degree safety problem into a standard stabilization problem, simplifying the design process for complex, nonlinear physical systems.
Key Foundational Work
Bridging the Gap between Optimal Trajectory Planning and Safety-Critical Control with Applications to Autonomous Vehicles (2020) — contributed 2 concept elements to this direction.
Control Barrier Functions: Theory and Applications (2019) — contributed 2 concept elements to this direction.