High-Level Multi-Robot Trajectory Planning And Spurious Behavior Detection This work has been supported by the ONR Global grant N62909-24-1-2081, Spanish project PID2024-159284NB-I00 funded by MCIN/AEI/10.13039/501100011033, by ERDF A way of making Europe and by the European Union NextGenerationEU/P
Fernando Salanova, Jesús Roche, Cristian Mahulea, Eduardo Montijano · 2025
Read High-Level Multi-Robot Trajectory Planning And Spurious Behavior Detection This work has been supported by the ONR Global grant N62909-24-1-2081, Spanish project PID2024-159284NB-I00 funded by MCIN/AEI/10.13039/501100011033, by ERDF A way of making Europe and by the European Union NextGenerationEU/P by Fernando Salanova, Jesús Roche, Cristian Mahulea, Eduardo Montijano (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-2510-17261.txt.
Introduction
Fernando Salanova and three co-authors catch wrong acts in multi-robot missions in this 2025 paper. Missions state goals as logic rules for time, place, and order. Robots can break those rules by wrong task order or late acts. The team builds fake mission data and a watcher to spot such breaks.
Their data frame nests robot nets inside a mission-wide net. A transformer watcher then sorts each path as sound or flawed. Tests flag waste in over nine in ten flawed paths. Core rule breaks are caught at a high rate as well. Weaker spots show in adaptive breaks tied to shifting limits.
Catalog record
| CREATOR | Fernando Salanova, Jesús Roche, Cristian Mahulea, Eduardo Montijano |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-17261
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
High-Level Multi-Robot Trajectory Planning And Spurious Behavior Detection This work has been supported by the ONR Global grant N62909-24-1-2081, Spanish project PID2024-159284NB-I00 funded by MCIN/AEI/10.13039/501100011033, by ERDF A way of making Europe and by the European Union NextGenerationEU/PRTR, and DGA T45-23R and T64-23R. Fernando Salanova, Jesús Roche, Cristian Mahulea, Eduardo Montijano Abstract The reliable execution of high-level missions in multi-robot systems with heterogeneous agents, requires robust methods for detecting spurious behaviors. In this paper, we address the challenge of identifying spurious executions of plans specified as a Linear Temporal Logic (LTL) formula, as incorrect task sequences, violations of spatial constraints, timing inconsistencies, or deviations from intended mission semantics. To tackle this, we introduce a structured data generation framework based on the Nets-within-Nets (NWN) paradigm, which coordinates robot actions with LTL-derived global mission specifications. We further propose a Transformer-based anomaly detection pipeline that classifies robot trajectories as normal or anomalous. Experimental evaluations show that our method achieves high accuracy (91.3%) in identifying execution inefficiencies, and demonstrates robust detection capabilities for core mission violations (88.3%) and constraint-based adaptive anomalies (66.8%). An ablation experiment of the embedding and architecture was carried out, obtaining successful results where our novel proposition performs better than simpler representations. I Introduction Multi-robot systems significantly enhance capabilities for complex missions such as surveillance, rescue, and area coverage by offering improved efficiency, robustness, and fault tolerance over single-agent systems [ 1 ] . Their ability to execute tasks in parallel and recover from individual failures makes them ideal for dynamic and high-stakes environments. However, coordinating multiple agents introduces new challenges, particularly in ensuring that the collective behavior of the team adheres to the intended mission. Spurious behaviors may emerge due to robot faults, delays, or incorrect task sequencing, which may not be immediately evident at the level of individual agents. Such deviations can compromise mission goals, introduce safety risks, or violate temporal constraints. Detecting these spurious trajectories is essential to guarantee mission integrity in multi-robot systems. Planning frameworks for robot teams often combine low-level motion planners (e.g., A*, RRT*, potential fields) [ 2 , 3 ] with high-level task models based on formal methods like Linear Temporal Logic (LTL), Petri nets (PNs), or hierarchical automata [ 4 , 5 ] . While low-level planners focus on collision-free execution in continuous space, high-level mission planners govern the temporal and logical structure of team behavior. The alignment between these layers is nontrivial, making it difficult to detect team spurious behaviors, given that they can arise even if each robot’s motion appears correct locally. To address this problem, the main contributions of this work are (i) the creation of a dataset including diverse multi-robot high-level mission behaviors and (ii) a learning-based pipeline that detects spurious multi-robot trajectories, those that violate either low-level constraints (e.g., forbidden zones, speed violations) or high-level mission semantics (e.g., incorrect task ordering, temporal failures). Unlike most prior work, which focuses primarily on local agent anomalies [ 6 , 7 ] , our approach analyzes entire team trajectories, enabling detection of subtle behavioral inconsistencies at multiple levels. Experiments in simulated environments show how our approach successfully identifies with high accuracy a wide range of spurious behaviors. II Related Work Prior work on planning under logical constraints includes Singh and Saha [ 8 ] , who propose online LTL planning with reactive re-planning. Shi et al. [ 9 ] use simulated annealing for Boolean constraints but lack temporal expressiveness. In contrast, our framework tightly integrates formal mission specifications, robot-level modeling, and cross-agent coordination. Anomaly detection in robotics often focuses on low-level faults. Nandakumar et al. [ 10 ] review anomaly types in Autonomous Robotic Missions (ARMs), but do not propose detection methods or high-level data representations. Häussermann et al. [ 6 ] use spatial-temporal SOMs for plan execution in single robots. Khatib et al. [ 11 ] apply ML-based methods to detect coordination anomalies in leader-follower teams, focusing on task-specific behaviors. Zhang et al. [ 12 ] explore statistical detection in large-scale trajectory data, though lacking structured mission modeling. Our approach addresses these limitations by detecting anomalies at the mission level using learned embeddings of structured symbolic trajectories. This enables end-to-end classification of multi-robot behaviors as mission-compliant or spurious, a capability underexplored in prior work. III Methodology Our methodology is structured into five main phases: Initially the robot team and mission is modeled using the NWN paradigm, explained in III-A ; subsequently using the model, we generate the trajectories as explained in III-B , then in III-C low-level timing estimation is obtained through VMAS simulations. Following this, an embedding of the generated data is computed to create a more representative structure of intrinsical behavioral patterns, as explained in III-D , and as the final step, in III-E a Transformer-based neural network is trained on these embedded trajectories to classify them. III-A Formal specification of multi-robot missions via NWNs Achieving complex cooperative behaviors in heterogeneous multi-robot systems requires sophisticated planning mechanisms based on discrete event systems abstractions. In this work, we adopt the Nets-within-Nets (NWN) paradigm, a hierarchical modeling formalism particularly well-suited for capturing high-level specifications and coordinating heterogeneous teams [ 13 ] . At its core, a NWN model of a multi-robot system consists of three main components, as illustrated in Fig. 1 . On the left, we observe the Specification Net , a PN system representing the global mission specification that must be fulfilled by the robot team. This mission is specified as a co-safe LTL formula over atomic propositions, (y_{1},y_{2},y_{3},\ldots) . Unlike general LTL formulas, which may require observing infinite traces, co-safe LTL formulas can be satisfied by a finite prefix of a trace. The specification net in Fig. 1 is defined by the LTL formula: \lozenge(y_{1}\land\lozenge(y_{2}\land\lozenge y_{3})), which imposes the robot team visiting y_{1} , y_{2} , and y_{3} in that specific order. Below on the figure, three PNs model the behaviors of three individual robots. Each robot has its own Robot Net , derived by abstracting the continuous environment into a discrete event system, for example, by a cell decomposition algorithm. When a robot occupies one region, the associated proposition is evaluated as true ( y_{1} in the figure), potentially enabling the firing of a transition in the specification net. Finally, top right of the figure shows the System Net . This high-level Petri net contains tokens that represent the previously described component nets: three tokens in place Rb , each of these tokens correspond to a robot net and another token in place Ms corresponding to the mission specification net. Also for each of these transitions, a different number of robots can move synchronously, e.g., one, two or three robots moving at the same time in the example net. Transitions in this system net are responsible for synchronizing the evolution of the robot models with the mission specification. The synchronization is managed via the Global Enabling Function (GEF), Java functions integrated into the system net (see [ 13 ] for further details). Fig. 1: Conceptual structure of the Nets-within-Nets framework. (i) Represents the Specification PN, in charge of ensuring a correct development of the mission. (ii) the System Net, that synchronizes the state evolution in respect to the mission and the robot actions. (iii) pictures the