Composing Control Barrier Functions for Complex Safety Specifications

Tamás G. Molnár, Aaron D. AmesIEEE Control. Syst. Lett 2023 · DOI

TL;DR

This paper introduces a novel framework for designing controllers that guarantee safety for complex systems with intricate safety requirements. They achieve this by developing a method to combine multiple safety constraints, expressed as Boolean logic (AND, OR, NOT), into a single, continuously differentiable control barrier function (CBF). This allows for more robust and implementable safety guarantees in sophisticated control scenarios.

What Problem Does This Paper Address?

Modern control systems are becoming increasingly complex, often requiring adherence to a multitude of safety constraints simultaneously. These constraints can be intertwined in sophisticated ways, going beyond simple "do this AND do that." For instance, a robot might need to avoid multiple obstacles (each with its own set of avoidance rules) while simultaneously navigating to a goal. Traditional control barrier functions (CBFs) are excellent for enforcing single safety constraints, but combining multiple CBFs to handle complex logical compositions (like "avoid obstacle A AND (avoid obstacle B OR avoid obstacle C)") has been challenging. Existing methods either result in discontinuous controllers (making them hard to implement) or struggle with arbitrary logical combinations. This paper aims to bridge this gap by providing a unified, smooth, and algorithmically derived CBF that can represent these complex, Boolean-composed safety specifications.

Methods

The core contribution of this paper is an algorithmic approach to construct a single, continuously differentiable control barrier function (CBF) that encapsulates complex safety specifications defined by Boolean combinations of individual safety constraints.

The authors build upon existing work by:

  • Boolean Logic for Safety Specifications: They represent safety specifications as sets in the state space, where each set corresponds to an individual safety constraint. These individual sets can then be combined using logical operations:
  • Union (OR): Represented by the maximum of the individual barrier functions.
  • Intersection (AND): Represented by the minimum of the individual barrier functions.
  • Complement (NOT): Represented by negating the barrier function.
  • Smooth Approximations: To overcome the non-smoothness of direct max/min operations (which would lead to non-smooth CBFs), they employ smooth approximations.
  • For unions (OR), they use a log-sum-exp function to smoothly approximate the maximum.
  • For intersections (AND), they use a similar log-sum-exp approach to smoothly approximate the minimum.
  • These smooth approximations result in a single, continuously differentiable CBF candidate.
  • Algorithmic Composition: The framework allows for arbitrary nesting of these Boolean operations. They propose a recursive method to build the final CBF by iteratively applying the smooth approximations for union and intersection.
  • Buffers for Tuning: The method incorporates tunable "buffers" that allow for adjusting the resulting safe set. This enables the approximation to either encapsulate the target safe set or lie within it, providing flexibility in the trade-off between conservatism and approximation accuracy.
  • Safety Filter Integration: The derived CBF is then used within a standard safety filter (a quadratic program) to modify a nominal controller, ensuring that the closed-loop system remains safe.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and demonstrates its effectiveness through simulated examples.

Key Insights and Ideas

  • Unified CBF for Complex Logic: The most significant insight is the ability to translate complex Boolean logic (AND, OR, NOT) of multiple safety constraints into a single, smooth CBF. This simplifies controller design and implementation compared to managing multiple discontinuous CBFs.
  • Smoothness is Key for Implementability: By using smooth approximations (log-sum-exp) for max/min operations, the paper ensures that the resulting CBF is continuously differentiable. This is crucial for generating smooth, implementable control inputs, avoiding the issues associated with non-smooth controllers.
  • Tunable Approximation: The introduction of buffers provides a practical mechanism to tune the conservatism of the safety guarantee. This allows engineers to balance strict adherence to the specified set with the ability to achieve performance objectives.
  • Algorithmic Scalability: The recursive nature of the proposed method suggests that it can handle arbitrarily complex compositions of safety specifications, making it scalable to highly intricate systems.

Key Results

The paper demonstrates the efficacy of its framework through several numerical examples:

  • Example 1 (Reach-Avoid with Obstacle): A rectangular agent needs to reach a goal while avoiding a single rectangular obstacle. The safe set is defined by the union of four constraints (top, bottom, left, right of the obstacle). The proposed CBF successfully guides the agent safely, and simulations show the controller input remains smooth.
  • Example 2 (Multiple Obstacles): This example extends the reach-avoid task to three obstacles. The safety specification involves ANDing the avoidance conditions for each obstacle, where each obstacle's avoidance is an OR of four constraints. The proposed method successfully generates a safe and smooth controller.
  • Example 3 (Road Network Navigation): The agent must navigate a road network, defined by inner and outer boundaries of multiple roads. The safety specification involves ANDing the inner and outer boundaries for each road, and then ORing these road safety conditions. The simulations show that the agent is kept safely on the road, and importantly, the safe set is attractive, meaning the agent can recover from unsafe initial conditions.

Limitations and Future Work

  • CBF Guarantee Not Explicitly Proven for All Cases: While the paper proposes a CBF candidate, Theorem 5 states that it approximates a nonsmooth CBF. The paper acknowledges that it is not guaranteed that the proposed smooth function is a CBF in all cases and that additional conditions (like those in Theorem 3) might be required. This means that while the approximation is good, formal CBF properties might need further verification.
  • Numerical Conditioning: The use of exponentials in the log-sum-exp approximation can lead to numerical issues if the smoothing parameter $\kappa$ is too large. The authors suggest scaling CBFs with class-K functions as a potential mitigation.
  • Computational Complexity: While the method is algorithmic, the computational cost of computing the smooth approximations and their derivatives might increase with the number of constraints and the depth of the logical composition.
  • Future Work: The authors suggest exploring other smooth approximations for max/min and investigating the conditions under which their proposed smooth function is guaranteed to be a CBF. They also mention that the framework could be extended to handle more complex temporal logic specifications.

Summary updated

BibTeX
@article{molnar2023composing,
  author = {Tamás G. Molnár and Aaron D. Ames},
  title = {{Composing Control Barrier Functions for Complex Safety Specifications}},
  journal = {IEEE Control. Syst. Lett},
  year = {2023},
  doi = {10.1109/lcsys.2023.3339719},
}
Figure