PRREACH: Probabilistic Risk Assessment Using Reachability for UAV Control
Nicole Fronda, Hariharan Narayanan, Sadia Afrin Ananna, Steven Weber, Houssam Abbas · 2025
Read PRREACH: Probabilistic Risk Assessment Using Reachability for UAV Control by Nicole Fronda, Hariharan Narayanan, Sadia Afrin Ananna, Steven Weber, Houssam Abbas (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-2509-04451.txt.
Introduction
Nicole Fronda and four co-authors design drone controllers that keep flight risk under a chosen limit in a 2025 research paper. Judging drone risk usually needs incident probabilities that scarce data cannot supply, and older frameworks offered no way to steer risk down. Their approach, PRReach, instead studies reachability: the full set of trajectories a drone could feasibly fly under its dynamics. Hazards are mapped from open spatial data, and risk is assessed over that whole trajectory set.
A control optimization then nudges the existing control law as little as possible while respecting the accepted risk threshold. Simulations on real-world data cut risk by up to about a quarter before flight and over half during flight.
Catalog record
| CREATOR | Nicole Fronda, Hariharan Narayanan, Sadia Afrin Ananna, Steven Weber, Houssam Abbas |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-04451
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
PRREACH: Probabilistic Risk Assessment Using Reachability for UAV Control Thanks: This work was partially supported by the FAA ASSURE Center of Excellence under projects A51 and A58, NSF Award 2118179, and the 2024 FAA Data Challenge . Thanks: 1 School of Electrical Engineering and Computer Science, Oregon State University, Corvallis, OR, USA {frondan,abbasho}@oregonstate.edu Thanks: 2 Department of Electrical Engineering and Computer Science, Philadelphia, PA, USA {hn384,sa3838,steven.weber}@drexel.edu Nicole Fronda, Hariharan Narayanan, Sadia Afrin Ananna, Steven Weber, Houssam Abbas Abstract We present a new approach for designing risk-bounded controllers for Uncrewed Aerial Vehicles (UAVs). Existing frameworks for assessing risk of UAV operations rely on knowing the conditional probability of an incident occurring given different causes. Limited data for computing these probabilities makes real-world implementation of these frameworks difficult. Furthermore, existing frameworks do not include control methods for risk mitigation. Our approach relies on UAV dynamics, and employs reachability analysis for a probabilistic risk assessment over all feasible UAV trajectories. We use this holistic risk assessment to formulate a control optimization problem that minimally changes a UAV’s existing control law to be bounded by an accepted risk threshold. We call our approach PRReach. Public and readily available UAV dynamics models and open source spatial data for mapping hazard outcomes enables practical implementation of PRReach for both offline pre-flight and online in-flight risk assessment and mitigation. We evaluate PRReach through simulation experiments on real-world data. Results show that PRReach controllers reduce risk by up to 24% offline, and up to 53% online from classical controllers. I Introduction Fig. 1 : On the left we see example UAV trajectories over a heatmap representing population density of city blocks in Philadelphia. The blue line shows the nominal trajectory over this area towards the target given by a red dot. The orange line shows the trajectory when the UAV experiences a deficient rotor, causing it to fly over a more densely populated area. The light blue rectangles show the reach sets of the UAV with the deficient rotor. On the right we see the same heatmap and nominal UAV trajectory. The orange line here shows the UAV trajectory using a PRReach controller, which directs the UAV away from the more densely populated regions. PRReach used the reach sets to evaluate risk over all feasible trajectories of the UAV, and minimally change the UAV controller to maintain a bounded level of risk. This results in the reduced risk evaluation from 0.79 to 0.61 between the left and right plots, at the cost of the UAV on the right being further from the target at the end of the time horizon. Best viewed in color. Uncrewed Aerial Vehicles (UAVs) have revolutionized various industries, from logistics and surveillance to environmental monitoring and emergency response. These operations often require a waiver [ 1 ] that provides a measure of risk of all potential hazard outcomes - an undesirable event such as collision with a person or building. A common measure of risk is the probability of such an event occurring times its severity. Current risk management frameworks for UAVs use Probabilistic Risk Assessments (PRA) to calculate risk using conditional probabilities of a hazard outcome given a hazard cause - a broken rotor, strong winds, sensor malfunctions, etc. However, these frameworks are severely constrained by lack of requisite data for computing these conditional probabilities, and are limited in their assessments and decisions. We propose a new approach for risk assessment and mitigation which, through reachability analysis, relates the dynamics of a UAV to the likelihood of hazard outcome as modeled by available spatial data of the UAV’s flight environment. We call our approach PRReach , for P robabilistic R isk Assessment Using Reach ability. PRReach can be applied by regulators evaluating risk of proposed UAV operations, and can aid operators in developing risk mitigation measures. The process for operators would be as follows: 1) If a hazard cause occurs before flight, use a precomputed PRReach controller to guarantee risk below a set threshold. Otherwise proceed with nominal flight controllers. 2) If no hazard cause is present initially but occurs later during flight either a) switch to the PRReach controller that was precomputed offline for the hazard cause, OR b) if time resources are available, compute a PRReach controller online . Figure 1 shows example trajectories from PRReach. Related Work Existing PRA frameworks for traditional aircraft follow the procedural requirements for executing safety risk management within the FAA [ 2 ] . Proposed PRA frameworks for UAVs follow these same requirements, including the framework most recently presented in FAA Assure Report 21 [ 3 ] . In addition to relying on limited historical UAV data for modeling risk, a limitation which PRReach does not have, this approach only considers categorical hazard causes - does wind exist or not. PRReach can incorporate continuous hazard causes, such as wind speed into its risk assessment. The approach in [ 3 ] is also a purely offline framework, meant to give recommendations for whether or not an operation can take flight. It does not provide a method for in-flight control. In contrast, PRReach allows for offline and online risk assessment, and produces risk-bounded controllers. Furthermore, [ 3 ] averages risk over all areas in the environment, and does not consider where the UAV can actually go according to its dynamics. Instead, PRReach uses reachability analysis to compute risk over all feasible trajectories for a risk evaluation. Reachability analysis is the computation of the set of future states of a dynamical system from some initial set of states. These sets are called reach sets . It has been widely used as a tool in motion planning and control for many different types of dynamical systems. Algorithms are available for computing the reach sets of discrete-time and continuous-time linear, non-linear, deterministic and stochastic systems, with many specializations for various cases. See [ 4 ] and [ 5 ] and the references therein for a background on the most common ways of computing reach sets and associated software tools. In this work, we use zonotopes to represent reach sets as they are closed under linear transformations. Conversion from zonotopes to other set representations is possible when necessary. For example, [ 6 ] describes scalable conversions between zontopes and ellipsoids. Existing control methods for robotic systems which use reachability include the works [ 7 , 8 , 9 ] . These works are similar to ours in that they use reach sets within an optimization problem to ultimately find trajectories that are bounded by some risk evaluation. However, PRReach differs in that it optimizes the generation of the reach sets themselves, and not over trajectories within the reach set. PRReach yields a risk-aware controller that can produce multiple trajectories for a UAV given different initial states, each guaranteed to not exceed some level of risk. As far as we know, this approach is the first that optimizes a controller in this way. Contributions In this paper we present: 1. A PRA method for computing risk over reach sets 2. A formulation for a risk-bounded control optimization 3. A solution to the optimization called PRReach, which we evaluate in simulation using real-world data and compare with a classical controller. II Preliminaries We