Following an environmentally constrained path in a timely fashion can be crucial for swarms in scenarios such as disaster response or emergency evacuation. In such situations, swarms must rapidly follow potentially challenging paths while not losing cohesion. We benchmark a robust, decentralized gradient-following behavior against varying path sinuosity, and swarm size. The swarm races to reach a minimum distance in a fixed time budget. We measure the success rate and the mean completion time. Our findings show that the algorithm allows the swarm to successfully reach the target line in the allotted time as long as the swarm size is relatively small. Furthermore, we find that the algorithm handles low and medium degrees of sinuosity well but struggles with high sinuosity, where the turns become too sharp. This work is the first to study swarm racing in constrained environments, revealing that “more is not always better”: larger swarms hinder rapid path traversal. This paves the way for future research on scale-invariant racing collective behaviors.
Fluid-like volumetric collective motion in drone swarms with navigation, obstacle avoidance, and formation maintenance remains challenging. Existing potential-field-based approaches often lack stability guarantees and are largely restricted to two-dimensional formations. This paper proposes Active Elastic Matter, an extension of the Active Elastic Sheet method that (i) generalizes the elastic interaction model to volumetric 3D formations and (ii) incorporates estimated relative velocities and accelerations using an Extended Kalman Filter. These predictive interaction terms improve stability and robustness in dynamic environments. Simulation results demonstrate that Active Elastic Matter enables 3D formations to navigate narrow passages and dynamic obstacles while maintaining higher order and lower entropy compared to Active Elastic Sheet. Real-world experiments with Crazyflie drones further validate improved stability during narrow passage navigation.
The tendency to favour members of one’s own group, also known as in-group bias, is ubiquitous in social mammals from rats to humans. This preference for in-group members (which often translates into aggression against out-group members) plays a major role in mammalian social life, and governs how individuals and groups perceive and behave toward others. Understanding how such social dynamics emerge, evolve, and impact behaviour is crucial for evolutionary theory and for the design of more mammal-like agents. Yet current swarm robots are typically extremely social and cooperative, and do not differentiate between individual robots. Here, we introduce two cognitive and social layers to a swarm engaged in a foraging task and a communication game: robot individuation (a partner-specific memory) and evolving sociality (a partner-specific tendency to interact based on previous interactions). We test whether social asymmetries such as in-group bias can spontaneously emerge from repeated local interactions without implicit group markers or pre-assigned identity tags, and study its strength and stability over varying conditions: cognitive (memory size and degree of decay), social (group size and innate sociality) and environmental (spread of resources). We show that implementing these two mammal-like features leads to the spontaneous emergence of in-group bias, which is robust across ecologies. These results shed light on the evolution of social asymmetries, and pave the way for modelling the complex social dynamics of mammals.
Tumblenauts are a swarm of minimalist, bacteria-inspired robots designed for collaborative inspection of pressurized microgravity habitats such as the International Space Station. Unlike current intra-vehicular robots that rely on complex, actuator-dense mechanisms for precise motion, the Tumblenauts use a stochastic run-and-tumble locomotion inspired by bacterial motility. This unique locomotion paradigm enables a simpler design, improves scalability, and greatly reduces actuation requirements, making the Tumblenauts among the smallest and least actuator-dense robots built for microgravity. In this paper, we present the design of the Tumblenaut, describe how it achieves run-and-tumble locomotion, and characterize its motion dynamics using Earth-based microgravity testbeds. Furthermore, using a data-driven simulation, we demonstrate how the Tumblenauts can perform a diverse set of inspection tasks by leveraging collective behavior. Specifically, we show that the swarm can collaboratively map environments, achieve directed navigation through a chemotaxis-inspired control mechanism, and make global inspection classification decisions by sharing information. As a new generation of space habitats is launched into orbit, we envision swarms of Tumblenauts run-and-tumbling within them, providing continuous monitoring and supporting the long-term sustainability of these stations.
Collective motion in swarm robotics is often designed from homogeneous, physics-inspired rules, yet real swarms must operate in confined arenas where boundaries reshape local interactions and where robots exhibit heterogeneous sensor and motor biases. We introduce an online, fully decentralized self-calibration framework that tunes microscopic control parameters so that the swarm reaches a desired macroscopic motility phase. Building on social learning, we add a fast-transmission mechanism that rapidly disseminates clearly better controllers through local interactions, complementing slower mutation–selection updates. Using a motility model combining alignment, density-dependent crowding, and collective U-turn propagation, we show in simulation that swarms self-calibrate from random initial behavior to reliably achieve several target phases (flocking, disordered gas, clustering, and gas–solid coexistence) in a disk arena. Fast transmission substantially accelerates convergence and improves robustness to heterogeneity, enabling rapid phase control without any central coordination.
This paper presents a constrained-control synthesis algorithm amenable for distributed multi-robot decision making. The proposed approach is based on a newly introduced controller selection method named analytic center selection which results in a dynamically defined controller that is able to ensures the satisfaction of state constraints. The controller synthesis lends itself to be distributed among a network of agents using state-of-the-art distributed optimization techniques. Resilience of the controller to imperfect communication channels among the robots is demonstrated in simulation.
In this paper, we present a fully distributed method that allows a group of magnetically connected modular robots to assess the mechanical stability of their structure. Stability is verified based on four mechanical phenomena: vertical and rotational sliding and rotational debonding for vertical and lateral connectors. We present a mechanical model applicable to the Blinky Blocks robot system. We also propose an efficient, fully distributed algorithm that runs simultaneously on all robots, enabling them to check global stability and detect the type and the breakage positions. Our algorithmic solution avoids the global resolution of the system and is effective for both free-loop and complex loop or multi-loop configurations; for the latter, it systematically enumerates all possible spanning trees and traverses each one to assess the stability of the modular structure. The approach was validated both in simulation with VisibleSim and on real Blinky Blocks hardware. Experiments demonstrate accurate and robust detection of all four breakage types in a wide variety of configurations, with low computational overhead and excellent scalability. These results confirm that distributed spanning-tree-based analysis provides an effective alternative to solving global equilibrium equations in modular robotics, enabling reliable structural integrity assessment even in large and highly connected assemblies.
We study whether leadership can arise in collectives of agents that follow identical interaction rules but differ slightly in perceptual or kinematic traits. Using a continuous-time flocking model with controlled micro-heterogeneity, we measure directional influence through a lagged-correlation network metric. Fully homogeneous groups show uniformly low influence, whereas even a small elite subset produces stronger leaders. Moderate heterogeneity amplifies leadership, while excessive heterogeneity reduces it. These results show that minimal parametric differences alone can generate emergent leaders in decentralized swarms.
This paper presents a distributed framework providing high-level positioning guidance for multi-UAV systems operating on a hemispherical domain. Unlike conventional coverage approaches restricted to planar environments, the proposed method optimizes a coverage objective directly on the hemisphere, enabling three-dimensional viewpoint coordination around a dynamic target. Each UAV computes its motion using only locally available data, while the target position and a predefined probability density function remain independent of team size. The framework is validated through simulations and real-world experiments, demonstrating scalability and robustness.
Understanding how neighbors influence each other in space and time is essential for explaining how collective behavior emerges. In this work, we present an information-theoretic approach that uses time-delayed mutual information to extract the optimal temporal influence delay and distance constrained transfer entropy to extract the spatial influence range. We first validate the approach with synthetic data generated by an agent based model with different interaction delays, spatial cutoffs, and noise levels, and we show that the method recovers the true temporal delay and spatial influence range. We then apply the approach to experiments in which a zebrafish follows a virtual fish performing controlled perturbations. The analysis reveals an interaction delay of about 600 milliseconds and a spatial interaction range between 3 and 6 body lengths, depending on how much information flows from the leader to the follower. These results demonstrate that information-theoretic tools can quantify the sensorimotor delays and spatial interaction ranges that shape social responses in animal collectives.
This paper addresses the problem of maintaining communication connectivity within a swarm of fixed-wing aerial vehicles operating under dynamic and range constraints. Fixed-wing platforms, while offering superior endurance and coverage capabilities compared to rotary-wing ones, pose additional challenges due to their nonholonomic dynamics and limited maneuverability. To tackle these challenges, we propose both centralized and distributed model predictive control formulations that explicitly integrate the algebraic connectivity of the inter-vehicle communication graph into the control framework. The resulting controllers allow each vehicle to anticipate connectivity degradation and adjust its trajectory proactively while pursuing observation and coverage objectives. The proposed approaches are benchmarked against heuristic and convex optimization-based controllers in a simulated multi-target surveillance scenario. Simulation results demonstrate that the distributed model predictive control scheme achieves comparable mission performance to centralized schemes while maintaining consistent network connectivity and robustness to target switches. These findings highlight the potential of predictive, connectivity-aware control for enabling scalable and resilient coordination of fixed-wing swarms in real-world applications.
This paper presents a heterogeneous viscoelastic cyber-physical swarm exploration algorithm for navigating robot groups in unknown, obstacle-filled environments. The framework integrates physical agents aware of a target coordinate with a surrounding layer of cyber agents that process environmental measurements and prevent collisions through viscoelastic interactions. Built on a structured, stable, and robust formulation, the method guarantees asymptotic stability of the collective motion and robustness to bounded perturbations. A hybrid control strategy enables agents to switch between normal operation and stagnation-recovery modes, helping the swarm avoid deadlock, maintain cohesion, and pass through narrow spaces without prior map knowledge. Monte Carlo simulations show that the algorithm consistently achieves alignment, stable cohesion, and collision-free exploration under sensing noise, varied swarm compositions, and different target coordinates in an unknown maze. The results demonstrate that the cyber-physical viscoelastic architecture provides adaptability and resilience, offering a scalable and analytically validated approach to autonomous exploration.
Information propagation in robot swarms is critical for coordinated behavior, since collective actions depend on the exchange of messages among robots. However, a predictive model linking the swarm parameters to the dynamics of information propagation has yet to be established. We introduce an epidemiology-inspired approach based on the Susceptible–Infected model to predict information propagation in mobile robot swarms executing random walks. The propagation rate is empirically related to robot density, communication range, and motion speed, yielding a predictive model that accurately reproduces the temporal evolution of the informed fraction across diverse swarm configurations.
In this work, we develop a user interface to manage a swarm of large fixed-wing Uncrewed Aerial Vehicles (UAVs) for firefighting applications through a user-centered design process. We conduct Wizard of Oz studies with nine firefighters and drone operators to collect end-user data on their ideal system for use in a wildfire scenario. This data is then translated into features and design drivers implemented in a prototype interface for swarm firefighting. The interface is evaluated through a usability study in which six participants provide feedback and score the interface, resulting in an overall usability rating of ’Good’ with a System Usability Scale (SUS) score of 76.25. We find that end users in this application are more concerned with producing the desired effect than with controlling individual UAVs. This finding provides a useful insight for the design of future multi-UAV systems, suggesting a reduced emphasis on micro-managing the fleet.
The collective perception scenario is an established swarm robotics task in which robots collectively infer a globally distributed environmental feature. The scenario is the most cited benchmark for collective decision-making in swarm robotics and has a 10-year history. Many variants of the original scenario have been proposed and studied. Many methods of collective decision-making have been tested against it. Given that the scenario was not initially intended as the defining benchmark, we summarize and analyze the literature and discuss the benefits and potential risks of a monoculture in benchmarking. We give a perspective on what could and should be improved in the future.
Robot faults are inevitable in swarm systems deployed in real-world applications, yet most fault mitigation approaches rely on explicit fault detection pipelines and hand-crafted recovery responses. We propose a novel Multi-Agent Reinforcement Learning (MARL) approach that learns a fault-robust controller by reactively mapping local state metrics directly to a set of predefined mitigation actions. We train a shared-parameter, recurrent MARL policy using Centralised Training with Decentralised Execution (CTDE) in a box transport task. In our approach, the same policy is used by both faulty and non-faulty robots, and in different fault conditions. Our learned policy demonstrates robustness to faults compared to a non-learning baseline, and scales to increasing numbers of faults not encountered during training. Results demonstrate that MARL provides a practical approach to robustness against faults in robot swarms, without relying on explicit fault detection.
We study the problem of deploying |R| mobile robots to maximize visibility coverage in a polygonal workspace while maintaining a line-of-sight communication path to a home location. The environment contains opaque obstacles, and robots become stationary sensing nodes once placed. Each node exposes visibility frontier windows—open segments of the current visibility boundary—and available robots evaluate these windows to select the point that yields the largest incremental visible area. Robots bid their predicted gain, and a decentralized auction installs the highest bidder as the next stationary node, preserving a connected “min-link” backbone. The process repeats until all robots are placed or the environment is fully covered. We present the algorithm, its complexity, and communication requirements. Experiments on synthetic maps show rapid, monotone growth of visible area and effective distributed decision-making.
Coordinating multiple drones for landing on vertiport-like infrastructures constitutes a general safety-critical swarm intelligence problem that combines discrete decision-making with domain-specific spatial occupancy constraints. Existing scheduling and control approaches often lack formal, executable correctness guarantees and remain loosely coupled to low-level control logic, making cyber-physical inconsistencies likely, especially when the number of drones increases. This work presents a formal and executable model for flight operation management of multiple tiny drones on multiple simplified vertiports based on bigraphical reactive systems. The proposed model captures both the spatial layout and the interactions among drones and the infrastructure, providing a unified understanding of a complex coordination problem. Compositionality enables the reconfiguration of takeoff and landing pad topologies, as well as fleet sizes, without altering the underlying rules. We employ model checking to generate the full state space from a set of initial states and verify safety invariants for 4 drones on a 3× 7 grid, and 3 drones on a 4× 5 grid. Landing success is formalized as a reachability condition. The compositional design of the specification facilitates reusability and modular verification, and the rule-based specification allows for execution of the traces of the proof.
We propose a paired physical and controller design for a swarm of simple robots attempting to collect objects and convey them to a goal region. We show that a robot body featuring a curved tail, combined with a simple state machine controller, enables scalable foraging even in a highly congested environment. Through numerical experiments, we compare our approach to a benchmark controller that uses omnidirectional movement and predictive collision avoidance. Our results demonstrate that the proposed approach cannot converge as quickly as the benchmark at low robot densities, but comes to surpass it at high densities. These findings highlight the value of integrating physical design with controller design to achieve scalability in robot swarms.
This paper presents a navigation function-based formation control methodology for autonomous surface vehicles (ASVs) operating in confined aquatic environments with currents and obstacles. Three methodological advances facilitate field deployment: (i) a differentiable approximation of the minimum distance function eliminating control chattering while preserving collision avoidance, (ii) environmental workspace boundary avoidance through virtual agents extending collision avoidance to environmental constraints, and (iii) a receding horizon waypoint strategy accommodating nonholonomic constraints and minimum speed thresholds. Unlike existing approaches requiring fixed spatial coordinates, the proposed framework allows robots to drift with currents while maintaining desired geometric patterns. Validation through numerical simulations and field experiments with Jaiabot micro-ASVs demonstrates robust performance despite GPS uncertainty, communication latency, and environmental disturbances.
