Abstract
Achieving simultaneous stabilization and safety in autonomous systems remains a significant challenge due to the potential for "deadlocks" where goal-seeking and obstacle-avoidance objectives conflict. This paper introduces the Nonsmooth Control Lyapunov Barrier Function (NCLBF), a mathematical framework that unifies these requirements using a "max" operator. By leveraging nonsmooth analysis and Mandalay derivatives, the proposed method eliminates local equilibria and ensures safe navigation toward a global goal in both linear and nonlinear systems.
Introduction
In control theory, two primary tools are used to govern autonomous behavior: Control Lyapunov Functions (CLFs) for reaching an equilibrium (stabilization) and Control Barrier Functions (CBFs) for maintaining safety (collision avoidance). Integrating these tools often leads to "gradient cancellation," where the system becomes trapped in a local equilibrium—effectively a "fake goal" where the push toward the target is perfectly balanced by the push away from an obstacle.
Existing solutions typically rely on Quadratic Programming (QP) or smooth blending functions. However, QPs can become infeasible in tight spaces, and smooth blending often creates the very local minima they seek to avoid. This paper addresses these shortcomings by proposing a nonsmooth approach that prioritizes the most critical constraint at any given moment.
Methodology: The Nonsmooth Advantage
The core innovation is the Nonsmooth Control Lyapunov Barrier Function (NCLBF). Instead of a smooth transition between objectives, the authors define the control function as:
$$V(x) = \max(L(x), B(x))$$
where $L(x)$ represents the goal-seeking Lyapunov function and $B(x)$ represents the obstacle-avoidance barrier function. This "max" operator creates a "kink" in the control surface, which is a deliberate design choice to prevent gradient cancellation.
As shown in the figure above, the state space is partitioned into specific regions. The system utilizes Mandalay derivatives—a generalized derivative computationally simpler than the Dini derivative—to handle the nonsmooth points. A piecewise feedback controller is then designed to switch logic based on whether the system is near the goal, near an obstacle, or on the boundary between the two.
Experimental Results
The NCLBF was validated through simulations of linear and nonlinear mechanical systems. Unlike traditional smooth CLBFs, which caused systems to stall when starting from specific initial positions, the NCLBF successfully guided the state to the origin in every test case.
The trajectories above demonstrate the system's ability to "weave" safely between bounded obstacles. In the nonlinear test case involving three distinct obstacles, the system maintained safety guarantees throughout the entire path. Furthermore, the Lyapunov value ($V(x)$) was shown to decrease monotonically over time, providing a rigorous proof of asymptotic stability.
Discussion and Future Work
A key insight of this research is that nonsmoothness is a powerful feature for avoiding local minima. By allowing for a non-differentiable control surface, the origin remains the only global minimum, effectively solving the deadlock problem for a wide class of Lipschitz control-affine nonlinear systems.
However, the current framework assumes that obstacles can be enclosed within "n-balls" (spheres). While mathematically sound, this can be conservative for elongated objects, potentially leading to inefficient paths. Future work will explore more complex Boolean constraints (e.g., "Avoid A OR B, but stay inside C") and investigate how these nonsmooth operations scale to high-dimensional systems, such as humanoid robots or multi-agent swarms.
Summary updated