Efficient Collision-Avoidance Constraints for Ellipsoidal Obstacles in Optimal Control: Application to Path-Following MPC and UAVs

David Leprich, Mario Rosenfelder, Markus Herrmann-Wicklmayr, Kathrin Flaßkamp, Peter Eberhard, Henrik Ebel · 2025

Read Efficient Collision-Avoidance Constraints for Ellipsoidal Obstacles in Optimal Control: Application to Path-Following MPC and UAVs by David Leprich, Mario Rosenfelder, Markus Herrmann-Wicklmayr, Kathrin Flaßkamp, Peter Eberhard, Henrik Ebel (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-2510-26531.txt.

Introduction

David Leprich, with five co-authors, helps drones steer around obstacles while following a set path. Their frame is model predictive path-following control, which plans a short stretch of motion, flies part of it, then plans again. Obstacles are treated as ellipsoids, stretched ball shapes, in three dimensions, and they may sit still or move.

Leprich and the team give a fast, smoothly varying test for whether the drone's path hits such an obstacle. A two-stage optimisation step calms numerical trouble that the control problem tends to cause. They show the method in simulation and in real flights with a Crazyflie quadrotor. The paper calls this the first hardware run of a controller of this kind for such drones in a three-dimensional task.

Catalog record

CREATORDavid Leprich, Mario Rosenfelder, Markus Herrmann-Wicklmayr, Kathrin Flaßkamp, Peter Eberhard, Henrik Ebel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-26531

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Efficient Collision-Avoidance Constraints for Ellipsoidal Obstacles in Optimal Control: Application to Path-Following MPC and UAVs David Leprich, Mario Rosenfelder, Markus Herrmann-Wicklmayr, Kathrin Flaßkamp, Peter Eberhard, Henrik Ebel Abstract This article proposes a modular optimal control framework for local three-dimensional ellipsoidal obstacle avoidance, exemplarily applied to model predictive path-following control. Static as well as moving obstacles are considered. Central to the approach is a computationally efficient and continuously differentiable condition for detecting collisions with ellipsoidal obstacles. A novel two-stage optimization approach mitigates numerical issues arising from the structure of the resulting optimal control problem. The effectiveness of the approach is demonstrated through simulations and real-world experiments with the Crazyflie 2.1 quadrotor. This represents the first hardware demonstration of an MPC controller of this kind for UAVs in a three-dimensional task. I INTRODUCTION In real-world robotic applications, collision avoidance is a crucial necessity for safe and autonomous operations. While numerous methods build upon artificial potential fields [ 1 ] and control barrier functions [ 2 ] , they typically modify the control inputs of a preexisting controller that does not explicitly account for obstacles [ 3 ] . Integrating collision avoidance into optimization-based motion planning and control is a non-trivial task. In particular, path-following methods are a common approach to effectively deal with a wide variety of robotic tasks, e.g., in surveillance, logistics, or agriculture [ 4 , 5 ] but are prone to collisions in complex environments. Defining a task by means of a geometric path is intuitive and straightforward for human operators, since, in comparison to defining time-dependent trajectories, a geometric path is time-independent and, therefore, considering the dynamics of the system is less important. Often, a path-planning algorithm is used to generate a geometric path for a specific task. Typically, in robotics practice, a higher-level path-planning algorithm, often operating on a purely geometric level, is responsible for providing such collision-free paths. However, in complex environments, the path-planning algorithm cannot always account for all relevant obstacles, particularly when there are (often smaller-scale) a-priori unknown obstacles discovered on-the-go, or when obstacles exhibit dynamic behavior. Even if the robot is equipped with onboard sensors capable of detecting such obstacles, continuously re-planning a global path in real-time often imposes prohibitive computational demands. Moreover, if robots additionally have kinematic constraints (e.g., nonholonomic constraints), it may not be possible to exactly follow a path purely planned based on geometrical considerations, so that a local dynamics-aware refinement of planned motions can be necessary, and should, ideally, be aware of obstacles locally. However, it is computationally non-trivial to realize a controller that is fast enough for closed-loop real-time operation while accounting for system dynamics and obstacles. Since obstacles can be understood as constraints in a robot’s configuration space, optimal control approaches like model predictive control (MPC) seem like a good candidate for the task as they can take into account both (often nonlinear) models of robot dynamics and constraints such as obstacles. However, it is non-trivial to include anti-collision constraints in optimal control problems (OCPs) in a differentiable and quick-to-evaluate manner, so that naïve implementations are typically slow and unreliable. This article overcomes this issue by building upon a recently proposed local obstacle-avoiding setpoint-tracking MPC controller [ 3 ] , similar to [ 6 ] . Central to the foundational approach is a computationally efficient and continuously differentiable condition for detecting collisions between ellipsoidal obstacles. The two main contributions of this article are the extension of the aforementioned works [ 3 , 6 ] to path tracking and to three-dimensional obstacle avoidance in practice, i.e., going beyond the planar case for the first time. In particular, static as well as moving obstacles are considered. Furthermore, numerical issues arising from the structure of the resulting optimal control problem (OCP) are mitigated by a novel two-stage optimization approach, helping to also keep computation times short enough in the three-dimensional case. Finally, the effectiveness of the approach is demonstrated in real-world experiments, utilizing the Crazyflie 2.1 quadrotor. To the best of the authors’ knowledge, this is the first time that an MPC controller of this kind is demonstrated in real-time in hardware experiments with an unmanned aerial vehicle (UAV) in a three-dimensional task. The article is organized as follows. First, an efficient test for the overlap of two ellipsoids is recapitulated in Section II , which serves as an algorithmic primitive for formulating this article’s obstacle-avoidance constraints. Subsequently, the collision-avoidance path-following MPC formulation together with a two-stage optimization approach is stated in Section III . Next, the dynamic model of the quadrotor used in the experiments is introduced in Section IV . Finally, the experimental results are presented in Section V and a conclusion is drawn in Section VI . II An Efficient Collision Test for Ellipsoidal Obstacles Detecting collisions between two objects is, in general, a non-trivial task that arises in various fields such as robotics, computer graphics, and physics simulations [ 7 ] . In this work, we restrict our attention to the case of ellipsoidal obstacles in three-dimensional space. However, the framework could be straightforwardly extended to higher-dimensional hyperellipsoids. Accordingly, both the robots and the obstacles are represented by ellipsoids \mathcal{E}(\bm{P},\bm{r})\subseteq\mathbb{R}^{3} , defined through a quadratic form as \mathcal{E}(\bm{P},\bm{r})=\{\bm{x}\in\mathbb{R}^{3}\,|\,(\bm{x}-\bm{r})^{\top}\bm{P}(\bm{x}-\bm{r})\leq 1\}, (1) where \bm{P}\in\mathbb{S}_{++}^{3} is a symmetric positive-definite matrix that characterizes the ellipsoid’s shape, size, and orientation, and \bm{r}\in\mathbb{R}^{3} specifies its center. Collision detection between a robot and an obstacle can then be formulated as checking whether the corresponding ellipsoidal sets, which may serve as outer approximations of the actual shapes, intersect. Note, that requiring \bm{P}\in\mathbb{S}_{++}^{3} might be overly restrictive, as the following approach can also be applied to positive semi-definite matrices \bm{P}\in\mathbb{S}_{+}^{3} , allowing obstacles to be defined by degenerate ellipsoids [ 8 ] , such as cylinders. This is practically relevant, for example, if a UAV is not allowed to fly over certain areas. A computationally efficient approach for testing for intersections of ellipsoidal sets, originally introduced in [ 8 , 9 ] and recently applied in two-dimensional collision avoidance for mobile robots in [ 3 ] , is summarized in the following. Given are two ellipsoids \mathcal{E}_{\mathrm{A}}(\bm{A},\bm{v}) and \mathcal{E}_{\mathrm{B}}(\bm{B},\bm{w}) with \bm{A},\bm{B}\in\mathbb{S}_{++}^{3} and \bm{v},\bm{w}\in\mathbb{R}^{3} . Testing for the intersection of \mathcal{E}_{\mathrm{A}} and \mathcal{E}_{\mathrm{B}} relies on constructing an auxiliary ellipsoid \mathcal{E}_{\lambda} obtained as a convex combination of \bm{A} and \bm{B} . Specifically, one defines \displaystyle\hskip-14.22636pt\mathcal{E}_{\lambda}\coloneqq \displaystyle\{\bm{x}\in\mathbb{R}^{3}\mid(\bm{x}-\bm{m}_{\lambda})^{\mkern-1.5mu\mathsf{T}}\bm{E}_{\lambda}(\bm{x}-\bm{m}_{\lambda})\leq K(\lambda)\}, (2a) \displaystyle\bm{E}_{\lambda} \displaystyle\coloneqq\lambda\bm{A}+(1-\lambda)\bm{B},\quad\lambda\in[0,1], (2b) \displaystyle\bm{m}_{\lambda} \displaystyle\coloneqq\bm{E}_{\lambda}^{-1}\big(\lambda\bm{A}\bm{v}+(1-\lambda)\bm{B}\bm{w}\big), (2c) \displaystyle\hskip-14.22636ptK(\lambda) \displaystyle\coloneqq 1-\lambda\bm{v}^{\mkern-1.5mu\mathsf{T}}\bm{A}\bm{v}-(1-\lambda)\bm{w}^{\mkern-1.5mu\mathsf{T}}\bm{B}\bm{w}+\bm{m}_{\lambda}^{\mkern-1.5mu\mathsf{T}}\bm{E}_{\lambda}\bm{m}_{\lambda}, (2d) where \mathcal{E}_{\lambda} always satisfies the containment property (\mathcal{E}_{\mathrm{A}}\cap\mathcal{E}_{\mathrm{B}})\subseteq\mathcal{E}_{\lambda}\subseteq(\mathcal{E}_{\mathrm{A}}\cup\mathcal{E}_{\mathrm{B}}). (3) The auxiliary ellipsoid \mathcal{E}_{\lambda} is always contained within the union of \mathcal{E}_{\mathrm{A}} and \mathcal{E}_{\mathrm{B}} , whereas the intersection of \mathcal{E}_{\mathrm{A}} and \mathcal{E}_{\mathrm{B}} is always contained within \mathcal{E}_{\lambda} , i.e., for every \lambda\in[0,1] , see the proofs in [ 9 ] . Note that, whereas the union of \mathcal{E}_{\mathrm{A}} and \mathcal{E}_{\mathrm{B}} is never empty, the intersection is, by definition, empty if and only if no collision occurs. Therefore, one possibility to enforce collision avoidance, i.e., \mathcal{E}_{\mathrm{A}}\cap\mathcal{E}_{\mathrm{B}}=\emptyset , is to ensure that it holds that \mathcal{E}_{\lambda}=\emptyset for an arbitrary \lambda\in[0,1] . The set \mathcal{E}_{\lambda} is empty if and only if K(\lambda) is negative for the specified \lambda as \bm{E}_{\lambda} is always positive-definite as the sum of two positive-definite matrices, leading to an empty set in the definition of the ellipsoid in (

Read the complete text