When designing autonomous systems—whether self-driving cars, quadrotors, or robotic arms—safety is paramount. We want these systems to achieve their goals while strictly avoiding unsafe regions of their state space. In modern control theory, Control Barrier Functions (CBFs) have emerged as a powerful tool to guarantee safety by acting as real-time safety filters.
However, a fundamental mathematical mismatch often arises between our safety goals and physical hardware: high relative degree.
The Core Challenge: Delayed Control Influence
To understand relative degree, consider a simple physical intuition. Suppose you are driving a car toward a wall, and your safety constraint is simple: do not hit the wall. Your constraint is defined in terms of position. However, your control input (the gas or brake pedal) directly changes the car's acceleration, not its position.
Mathematically, if you differentiate the position constraint once, you get velocity. The control input (acceleration) still does not appear. You must differentiate a second time for the control input to explicitly show up in the equations. The number of times you must differentiate a safety constraint before the control input explicitly appears is called the relative degree of the system.
As surveyed by Aaron D. Ames et al. in Control Barrier Functions: Theory and Applications, many real-world safety requirements involve a high relative degree. This creates a delayed control influence: an action taken right now does not immediately alter the system's proximity to the safety boundary; its effects take time to propagate through layers of integration (acceleration $\rightarrow$ velocity $\rightarrow$ position).
Why Standard Barrier Functions Fail
A standard, first-order Control Barrier Function assumes a relative degree of one. It works by ensuring that as a system approaches the boundary of a safe set, the control input can immediately steer the system's derivative away from the danger zone.
When the relative degree is higher than one, this approach breaks down. If you attempt to apply a standard CBF to a position constraint while controlling acceleration, the control input $u$ vanishes from the first derivative of the safety function. Because the input has no immediate mathematical leverage over the first derivative, a standard safety filter cannot choose an appropriate control action to prevent a future violation. As noted by Wei Xiao et al. in Control Barrier Functions for Systems with High Relative Degree, this mismatch requires complex recursive differentiation to ensure that control actions can effectively maintain system safety.
If a controller only looks at the immediate boundary, it might drive the system toward the wall at a high velocity. By the time the system is right next to the wall, even maximum braking acceleration cannot prevent a crash due to the system's momentum. The controller is mathematically blind to the delayed consequences of its current actions.
The Solution: High-Order Control Barrier Functions
To overcome this, the authors of Control Barrier Functions for Systems with High Relative Degree introduced High Order Control Barrier Functions (HOCBFs).
Instead of looking only at the final safety boundary, HOCBFs construct a sequence of nested, lower-order safe sets. Each layer in the sequence accounts for a higher derivative of the safety constraint. For a system with a relative degree of $m$, the framework defines a series of functions $\psi_0, \psi_1, \dots, \psi_m$, where each function constrains the derivative of the previous one:
$$\psi_k(x) = \dot{\psi}_{k-1}(x) + \alpha_k(\psi_{k-1}(x))$$
By the time the differentiation reaches the $m$-th layer, the control input $u$ finally appears explicitly. By enforcing a constraint on this final layer, the controller implicitly accounts for all the lower-order dynamics (like velocity and acceleration).
This mathematical formulation builds a "buffer zone" or a safe tube. If the system is moving fast toward an obstacle, the higher-order constraints force the controller to start braking early, long before the physical position boundary is breached. This ensures forward invariance—meaning if the system starts safe, it is mathematically guaranteed to stay safe.
Managing Real-World Complexities
While the mathematical foundation of HOCBFs solves the relative degree problem in ideal scenarios, real-world robotic deployment introduces several distinct challenges.
1. Heterogeneous and Mixed Relative Degrees
In complex systems, different actuators might influence safety constraints at different rates. For instance, in an autonomous vehicle, steering might affect lateral position at a different relative degree than acceleration affects longitudinal position.
Traditional safety filters often default to the input with the lowest relative degree, which effectively throws away available control authority. To address this, Wei Xiao et al. developed a framework in Control Barrier Functions for Systems with Multiple Control Inputs to handle heterogeneous relative degrees, allowing multiple actuators to be utilized optimally. Similarly, M. Cohen et al. proposed a constructive method using partial feedback linearization in Constructive Safety-Critical Control: Synthesizing Control Barrier Functions for Partially Feedback Linearizable Systems to handle systems where output components exhibit mixed relative degrees.
2. Model Uncertainty and Learning
HOCBFs rely heavily on knowing the exact model of the system to compute the required derivatives. If the physics model is inaccurate, the safety guarantees can fail.
To mitigate this, researchers have integrated machine learning to estimate unknown dynamics online. For example, Yu Zhang et al. introduced a framework combining sparse Gaussian Processes with HOCBFs in Real-Time Adaptive Safety-Critical Control with Gaussian Processes in High-Order Uncertain Models to handle model inaccuracies in real time. Additionally, Mohammad Javad Khojasteh et al. explored this in Probabilistic Safety Constraints for Learned High Relative Degree System Dynamics, providing probabilistic safety buffers that adapt as the system learns its own dynamics from noisy data streams.
3. Unmatched Disturbances
Another critical vulnerability is external disturbances (like wind gusts on a drone) that enter the system through a different channel than the control inputs.
When a disturbance affects the system at a lower relative degree than the control input, the disturbance begins to degrade safety before the controller has the mathematical leverage to counteract it. As explored by Jun Yang in Safety-Critical Disturbance Rejection Control of Nonlinear Systems With Unmatched Disturbances and Vrushabh Zinage in Disturbance Observer-based Robust Integral Control Barrier Functions for Nonlinear Systems with High Relative Degree, incorporating high-order disturbance observers allows the safety filter to anticipate these unmatched forces and proactively adjust the control boundaries.
Alternative Perspectives: Barrier States
While HOCBFs solve the problem by recursively differentiating the safety constraint, an alternative paradigm involves reshaping the system's state space directly.
Instead of treating safety as an external filter applied to a controller, the authors of Safety Embedded Control of Nonlinear Systems via Barrier States proposed embedding Barrier States (BaS) directly into the system's definition of stability. By transforming the state space such that the safety boundaries are pushed to infinity, safety and tracking goals are unified into a single control objective. As demonstrated by Maitham F. AL-Sunni et al. in Safety Embedded Adaptive Control Using Barrier States, this approach bypasses some of the recursive complexities of traditional HOCBFs, providing a distinct mathematical avenue for ensuring safety in high relative degree systems under uncertainty.
Key Foundational Work
Input-to-State Safety With Control Barrier Functions (2018) — contributed 2 concept elements to this direction.
Control Barrier Functions for Systems with High Relative Degree (2019) — contributed 1 concept elements to this direction.
Self-triggered Control for Safety Critical Systems using Control Barrier Functions (2019) — contributed 1 concept elements to this direction.