Dynamic Safety in Complex Environments: Synthesizing Safety Filters with Poisson’s Equation
Gilbert Bahati, Ryan M. Bena, Aaron D. Ames · 2025
Read Dynamic Safety in Complex Environments: Synthesizing Safety Filters with Poisson’s Equation by Gilbert Bahati, Ryan M. Bena, Aaron D. Ames (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-06794.txt.
Introduction
Gilbert Bahati, Ryan M. Bena, and Aaron D. Ames present in 2025 a method for keeping robots safe in places that change. Robots need a clear safe set, a zone where their actions stay safe, even as obstacles move. The authors build that set from sensor maps by solving a classic equation from physics.
Solving the equation gives a smooth field that guides the robot away from danger, and a safety function that marks the safe set. That function can then filter control actions so unsafe moves are blocked. Tests on four legged and humanoid robots in crowded, changing spaces show the method works in real time.
Catalog record
| CREATOR | Gilbert Bahati, Ryan M. Bena, Aaron D. Ames |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 81 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-06794
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Dynamic Safety in Complex Environments: Synthesizing Safety Filters with Poisson’s Equation Gilbert Bahati, Ryan M. Bena, Aaron D. Ames Abstract Synthesizing safe sets for robotic systems operating in complex and dynamically changing environments is a challenging problem. Solving this problem can enable the construction of safety filters that guarantee safe control actions—most notably by employing Control Barrier Functions (CBFs). This paper presents an algorithm for generating safe sets from perception data by leveraging elliptic partial differential equations, specifically Poisson’s equation. Given a local occupancy map, we solve Poisson’s equation subject to Dirichlet boundary conditions, with a novel forcing function. Specifically, we design a smooth guidance vector field, which encodes gradient information required for safety. The result is a variational problem for which the unique minimizer—a safety function —characterizes the safe set. After establishing our theoretical result, we illustrate how safety functions can be used in CBF-based safety filtering. The real-time utility of our synthesis method is highlighted through hardware demonstrations on quadruped and humanoid robots navigating dynamically changing obstacle-filled environments. I Introduction Deploying robotic systems in real-world environments autonomously requires that they operate in complex, dynamic environments while avoiding collisions with multiple objects of arbitrary geometry. Achieving this level of dynamic safety necessitates a quantifiable description of the safety requirement, i.e. a functional representation of the environment via a safety constraint. Additionally, this representation must be integrated with the dynamics of the system to produce safe inputs, i.e., inputs that ensure satisfaction of the safety requirements. To achieve this, numerous methods for enforcing dynamic safety constraints have been investigated, including Hamilton-Jacobi reachability [ 1 , 2 , 3 ] , Model Predictive Control [ 4 , 5 ] , Artificial Potential Fields (APFs) [ 6 , 7 ] , and Control Barrier Functions (CBFs) [ 8 ] . This paper seeks to fuse the synthesis of safety constraints with controllers that yield safe behaviors, coupling functional representations of dynamic environments with the generation of safe controllers. Safety constraints are typically synthesized using heuristic approaches for simple environments, or through general methods that are non-constructive. While heuristic methods work for environments with simple geometries [ 9 , 10 ] , they struggle when generalized to more complex environments. Alternatively, for more complex geometries, signed distance functions (SDFs) [ 11 , 12 , 13 ] have been proposed as a method for generating safety descriptors in the context of collision avoidance. However, SDFs possess gradient discontinuities which present challenges in synthesizing safe controllers. Therefore, the synthesis of safety constraints is a challenging problem in its own right—made even more challenging by the need to couple safety constraints with system dynamics. Fig. 1: Safe set synthesis from perception data via Poisson’s equation. Hardware experimental footage: https://youtu.be/fBRdkAJGixI . Given a description of the environment in the context of a goal location subject to obstacles, a popular method for enforcing safe behavior is the APF. Originally introduced in [ 6 ] , APFs are a heuristic technique for achieving dynamic safety by encoding “repulsive” gradients on obstacle surfaces and “attractive” gradients to a desired goal. This coupling between the dynamics of the system (via gradients) and specifications (via attractive and repulsive potentials) allows for the problem to be framed as one of constructing an APF. One approach to constructing APFs is to frame them as solutions to elliptic partial differential equations (PDEs) [ 14 , 15 ] . This has been employed in the context of navigation and path planning [ 16 , 17 , 18 ] and is useful for complex geometry [ 19 ] . Despite the success of APFs, this method assumes single integrator dynamics, i.e., a kinematic model. Additionally, APFs couple goal-reaching (stability) and collision avoidance (safety), limiting their ability to provide a description of safety that is independent of control objectives. A decoupled approach to synthesizing safe controllers for general nonlinear systems is through CBFs [ 20 , 8 ] . In this framework, safety constraints yield a safe set defined as the superlevel set of a function, and this function is a CBF if there exists inputs that render the safe set forward invariant. CBFs have been shown to generalize APFs [ 7 ] , in that they decouple safety from stability, and apply to general nonlinear systems. Specifically, given a CBF, one can synthesize a safety filter [ 21 , 22 ] framed as a quadratic program with the stability objectives as a cost and safety objective as a constraint. Methods such as [ 23 , 24 , 25 ] offer a systematic approach for constructing CBF-based safety filters for robotic systems via reduced-order models with real-time realizations on hardware. However, for collision avoidance in complex environments, the constructive design of functions encoding safety objectives, i.e., the synthesis of safe sets and the corresponding CBFs, remains a challenging task. In this paper, we show that elliptic PDEs can be used to synthesize safety functions —functional representations of complex environments that characterize safety. We demonstrate that the process of constructing safe sets can be formulated as solving a Dirichlet problem for Poisson’s equation. In particular, given an occupancy map, we solve Poisson’s equation subject to Dirichlet boundary conditions, where the solution, a safety function, characterizes a safe set, while the boundary conditions encode desired level-set values on obstacle surfaces. We propose several methods for constructing the forcing function within Poisson’s equation, including an average flux method and a guidance field method [ 26 ] that provides additional degrees of freedom for defining safety. The solution has desirable differentiability (i.e., regularity) properties, which are critical in enabling their use in the synthesis of safety filters. The key observation of this paper is that safety functions obtained from Poisson’s equation can be used to synthesize CBFs and, therefore, safety filters. The main contributions are threefold: (1) we present a constructive way of generating safe sets for complex environments from perception data via Poisson’s equation, (2) we illustrate and prove how the resulting safety functions can be used to synthesize CBF-based safety filters, and (3) we demonstrate the real-time efficiency of our approach with hardware experiments on quadruped and humanoid robots in static and dynamically changing environments. Importantly, we show that safety function synthesis (1) can be done in real-time via perception data to enable safe behaviors (2) on highly dynamic robotic systems (3). Notation: • A continuous function \gamma:\mathbb{R}\rightarrow\mathbb{R} is an extended class \mathcal{K} , \mathcal{K}^{e}_{\infty} , (\gamma\in\mathcal{K}^{e}_{\infty}) if \gamma is monotonically increasing, \gamma(0)=0 , \lim_{s\rightarrow\infty}\gamma(s)=\infty , and \lim_{s\rightarrow-\infty}\gamma(s)=-\infty . • For a function w:\mathbb{R}^{n}\rightarrow\mathbb{R} where \mathbf{x}\mapsto w(\mathbf{x})=w(x_{1},\cdots,x_{n}) , let Dw denote the gradient and D^{2}w denote the Hessian. More generally, for k\in\mathbb{N}_{0} , D^{k}w is the collection of all partial derivatives of order k . Given a multi-index \xi=(\xi_{1},\cdots,\xi_{n})\in\mathbb{N}_{0}^{n} with |\xi|=k : D^{k}w\coloneqq\big\{D^{\xi}w\,\big|\,|\xi|=k\big\},\quad D^{\xi}w=\frac{\partial^{|\xi|}w}{\partial x_{1}^{\xi_{1}}\cdots\partial x_{n}^{\xi_{n}}}. • \Omega is an open, bounded and connected set with smooth boundary \partial\Omega such that \overline{\Omega}=\Omega\cup\partial\Omega is the closure of \Omega . • C^{k}(\Omega) is the set of functions: \{w:\Omega\rightarrow\mathbb{R}\,\big|\, w is k -times continuously differentiable \}. • C^{k}(\overline{\Omega})=\{w\in C^{k}(\Omega)\,|\, D^{\xi}w is uniformly continuous on bounded subsets of \Omega for all |\xi|\leq k }, that is, D^{\xi}w continuously extends to \overline{\Omega} . • C^{k}(\Omega;\mathbb{R}_{\geq 0}) = \{w:\Omega\rightarrow\mathbb{R}_{\geq 0}\,\big|\, w is k -times continuously differentiable