Lukas Brunke,Siqi Zhou,Angela P. Schoellig
Abstract:Control barrier function (CBF)-based safety filters are used to certify and modify potentially unsafe control inputs to a system such as those provided by a reinforcement learning agent or a non-expert user. In this context, safety is defined as the satisfaction of state constraints. Originally designed for continuous-time systems, CBF safety filters typically assume that the system's relative degree is well-defined and is constant across the domain; however, this assumption is restrictive and rarely verified -- even linear system dynamics with a quadratic CBF candidate may not satisfy this assumption. In real-world applications, continuous-time CBF safety filters are implemented in discrete time, exacerbating issues related to violating the condition on the relative degree. These violations can lead to the safety filter being unconstrained (any control input may be certified) for a finite time interval and result in chattering issues and constraint violations. We propose an alternative formulation to address these challenges. Specifically, we present a theoretically sound method that employs multiple CBFs to generate bounded control inputs at each state within the safe set, thereby preventing incorrect certification of arbitrary control inputs. Using this approach, we derive conditions on the maximum sampling time to ensure safety in discrete-time implementations. We demonstrate the effectiveness of our proposed method through simulations and real-world quadrotor experiments, successfully preventing chattering and constraint violations. Finally, we discuss the implications of violating the relative degree condition on CBF synthesis and learning-based CBF methods.
What problem does this paper attempt to address?
### Problems the paper attempts to solve
This paper aims to solve the problems encountered in the practical application of the Control Barrier Function (CBF) safety filter. Specifically, the paper focuses on the following key issues:
1. **Non - satisfaction of the relative - degree assumption**:
- The CBF safety filter usually assumes that the relative degree of the system is well - defined and constant throughout the domain. However, this assumption often does not hold in practical applications. Even for linear system dynamics and quadratic CBF candidate functions, this assumption may not be satisfied.
- The non - satisfaction of this assumption can lead to an unconstrained safety filter (i.e., any control input may be certified), thereby causing chattering and constraint violations.
2. **Problems in discrete - time implementation**:
- Although the CBF safety filter is usually designed in continuous - time form, it needs to be implemented at discrete - time steps in practical applications. This will exacerbate the problems caused by the non - satisfaction of the relative - degree assumption.
- Discrete - time implementation may lead to the application of unsafe control inputs within a finite time interval, further increasing the system's insecurity.
### Solutions
To address the above problems, the paper proposes the following solutions:
1. **Multi - CBF method**:
- Multiple CBFs are introduced to generate bounded control inputs at each state, thereby preventing the incorrect certification of arbitrary control inputs.
- This method ensures that, by using multiple CBFs at each state point, a safe control input can be generated even when the relative degree changes.
2. **Conditions for the maximum sampling time**:
- The conditions for the maximum sampling time to ensure safety in discrete - time implementation are derived. These conditions ensure that the safety filter implemented at discrete - time steps can still provide continuous - time safety guarantees.
3. **Theoretical and experimental verification**:
- The effectiveness of the proposed method is verified through simulation and actual quadrotor UAV experiments. The experimental results show that this method successfully prevents chattering and constraint violations.
### Main contributions
1. **Defined CBF in systems with changing relative degrees**:
- It is pointed out that changes in the relative degree may lead to an unconstrained safety filter, which allows any control input to be certified.
2. **Introduced the multi - CBF method**:
- A method using multiple CBFs is proposed to prevent unconstrained safety filters, reduce chattering, and enhance the security of discrete - time implementation.
3. **Provided theoretical guarantees for the minimum sampling time**:
- Using the multi - CBF method, theoretical guarantees for the minimum sampling time to ensure safety in discrete - time implementation are provided.
4. **Experimental verification**:
- Through simulation and actual quadrotor UAV experiments, the effectiveness of the proposed method is demonstrated.
### Conclusion
By introducing the multi - CBF method and deriving the conditions for the maximum sampling time, this paper solves the problems encountered by the CBF safety filter when the relative - degree assumption is not satisfied and in discrete - time implementation. The experimental results show that this method has high effectiveness and reliability in practical applications.