Sino-German Workshop 2026, Konstanz, August 3-4

Collective behaviour emerges when individuals respond to one another: animals form flocks, cells coordinate their activity, diseases spread through populations, and robots organize themselves into teams. Most network models describe these systems through pairwise interactions. In many real situations, however, an individual responds to several others simultaneously, and the resulting group interaction cannot be reduced to a collection of independent pairwise links.

Higher-order networks, modelled using hypergraphs or simplicial complexes, offer a systematic way to represent such multi-party interactions. In this talk, I examine how higher-order interactions reshape collective dynamics. Using diffusion and spreading processes, I introduce analytical tools for studying stability, convergence, and amplification on these networks. I then discuss applications to epidemic dynamics, flocking and coordination in animal or robot populations, and strategic interactions in which individuals adapt their behaviour according to the choices of a group. These examples show that higher-order interactions can create qualitatively new collective phenomena and provide a common mathematical language for problems across biology, robotics, and network science.

Far-reaching perspectives in swarm robotics consider robots that are minimalist in their sensing, communication and computation, but are deployed in thousands to collaborate towards the accomplishment of tasks distributed in space and time. Generally speaking, future robot swarms might face harsh operating conditions where little communication is possible and no external infrastructure is available. These robots might not be able to efficiently communicate with a central controller and might not have the perceptual and computational abilities to self-localise or precisely plan their movements. It is therefore necessary to study collaborative strategies that do not rely on complex control and interaction rules. In this talk, I will present studies about minimalist approaches to collective behaviours based on random walks and simple interactions. I will introduce the concept of Adaptive Random Walks as a tool to design simple emergent behaviours like team formation and aggregation in minimalist robot swarms, showing how heterogeneity in the swarm can be beneficial to improve efficiency while maintaining the complexity low.

Biomimetics and Bioinspired soft robotics transfers the functions and principles nature into technical applications, enables us to create novel functions for soft robotics system like immanent to materials system such as embodied collective intelligence. Utilizing pneumatic logic circuits and transferring animal motion principles and gaits enabled the development of electronic free soft autonomous walkers. By integrating soft sensors, we provide soft robots with an intelligent decision-making system capable of processing information and directing its locomotion. The combination of bioinspired design principles and advanced fabrication techniques opens up new avenues for the development of autonomous systems.

Understanding the evolved mechanisms that govern collective behavior is a central challenge in studying animal societies. This talk presents two projects aimed at reverse-engineering the algorithmic basis of social interactions, with the additional objective of leveraging the biological findings for robotic systems. We explore these principles across two distinct systems: collective thermal soaring in birds and schooling in zebrafish.

First, we examine the challenging physical environment of atmospheric convection. We introduce a reverse-engineering framework that treats bird flocks as a distributed sensor array, allowing us to map the intricate, turbulent structure of thermal updrafts using only collective flight trajectories.

These data provide a unique environment for training Reinforcement Learning (RL) agents. We demonstrate that an AI-based autopilot can learn to soar with bird-like efficiency by utilizing nearby agents—both biological and artificial—as social cues. This highlights a future of bio-hybrid harmony, where artificial systems enhance their own performance by paying attention to the collective sensing of their biological counterparts.

Second, we investigate the fine-scale mechanisms of sensory-motor control using high-fidelity virtual reality (VR). By networking real fish with virtual robot agents, we deconstruct how zebrafish translate visual input into movement decisions. We show that schooling is driven by a robust, egocentric control law which accounts for all key features of pursuit behavior. This simple, evolved law requires minimal tuning when implemented in autonomous vehicles, achieving near-optimal performance across terrestrial, airborne, and aquatic platforms.

Through these examples of large-scale environmental sensing and fine-scale sensory input, we show that understanding the decision-making of animal groups provides more than mere bio-inspiration for robotics. Rather, it enables the creation of artificial systems capable of working harmoniously alongside animal societies, fostering a new level of cooperation between technology and the natural world.

Animals still far outpace legged robots in walking capability. Part of this success is due to the large amount of multimodal, distributed sensory feedback available to the nervous system to inform behavior, such as mechanosensors providing information on limb loading. Endowing robots with similar sensory capacity stands to improve their walking capability; however, we presently have a limited understanding of specific information encoded by biological sensors and how this information is used by the nervous system to drive behavior. To explore these questions, we have used the robot Drosophibot II, a biomimetic model of an adult Drosophila melanogaster fruit fly, to model insect-like mechanosensation during free walking. The robot was outfitted with strain gauges in similar locations and orientations as fields of campaniform sensilla strain sensors in the fly. We then recorded the responses of these sensors as the robot walked across different terrains. We found that this distributed, low-dimensionality feedback can robustly detect aspects of the terrain such as incline, as well as encode single-leg perturbations across multiple legs. In the future, this data can be paired with neurobiological studies in Drosophila to produce a mechanistic understanding of mechanosensation.

A shared goal of neuroscience and robotics is to understand how systems can be built to move effectively and autonomously through the world. However, state-of-the-art algorithms for selecting and executing limb-dependent behaviors in robots are still quite primitive compared with the neural controllers used by animals. To inform the design of artificial systems, we are investigating how the fly, Drosophila melanogaster, selects and controls its behaviors and how this process can be modulated by learning. I will discuss how we are pursuing this research program through a combination of experimentation and computational modeling.

All animals must make appropriate, but also flexible, foraging decisions, especially when facing starvation. Drosophila larvae need to consume enough nutrients, such as protein and carbohydrates, to be able to pupate and become a fly. Previous studies showed that larvae even perform cannibalistic behavior to reach this goal. We have now investigated whether differing social contexts or changes in internal states influence behavioral decisions to feed on conspecifics. We find that groups of fly larvae rarely use dead conspecifics as a food source. However, testing larvae as individuals enhances cannibalistic feeding. We additionally found that food deprivation enhances this feeding behavior in larval groups. Thus, larvae are attracted to dead conspecifics and engage in cannibalistic behavior when alone or when food-deprived. We have now started investigating the distinct underlying sensory and neuromodulatory mechanisms. Fly larvae sense conspecifics via multisensory cues. We further found that serotonin is required for food deprivation-dependent enhancement but not for social context modulation. Generally, this flexibility in feeding behavior enables the larva to explore a broader range of stimuli and to expand their feeding choices to overcome starvation.

Various underwater organisms have evolved over billions of years and possess excellent underwater living abilities. Inspired by this, we have developed various biomimetic driven underwater unmanned systems that can achieve various underwater motion capabilities. Inspired by the weak electric fish, an underwater electric field based sensing and communication system has been developed. Inspired by octopuses, we developed a soft bodied gripper that mimics octopuses. This report will systematically report the relevant research results.

Swarm robotics research has progressed a lot in the last decades. Most of these efforts, however, were driven through and showcased in the lab. Currently we observe the emergence of applications and more and more deployments outside of the lab. Pushing swarm robotics to a new 'swarm field robotics' is challenging and deployment is costly. Besides the popular research on flying drones, we need to implement progress also in other domains. In our own research, we develop a custom platform of a small autonomous sailboat that can operate and co-operate in teams on lakes. I'll present the hardware design, a theoretical result, and first results from the field. 

Compared with traditional propeller-driven underwater robots, robotic fish feature superior maneuverability and environmental adaptability, showing great potential for complex underwater perception tasks. However, the unsteady self-induced hydrodynamic signals generated by undulatory propulsion are strongly coupled with external flow disturbances, severely limiting the autonomous perception performance of artificial lateral line systems in free-swimming scenarios. This dissertation systematically addresses the key perception bottlenecks of free-swimming robotic fish. First, a data-theory hybrid interpretive framework based on Proper Orthogonal Decomposition is established to decouple and explain self-induced pressure signals, which is verified to be generalizable for diverse biomimetic models. Second, a robust motion state estimation method is proposed using modal dynamic characteristics, realizing accurate velocity and trajectory perception under variable swimming states and interactive flow interference. Third, a physical interpretable criterion for sensor array optimization is developed to achieve efficient and redundant-free sensor configuration. Finally, a bionic perception framework with self-motion signal compensation is constructed to eliminate self-induced interference, achieving precise near-field target localization. This study establishes a systematic hydrodynamic perception framework, providing theoretical support and engineering solutions for high-precision bionic underwater perception in complex flow environments.

How do animals make decisions? This question lies at the heart of neuroethology, yet much of what we know comes from behaviors studied under constrained laboratory conditions. While these approaches offer control, they often come at the cost of context, particularly for behaviors such as navigation, migration, and collective motion, which are shaped by the environments in which they unfold. Bridging this gap requires not only recreating aspects of the natural world, but accessing behavior as it actively occurs within it. Here we present MATREX, an open-source, modular virtual reality platform designed to bring experimental control into ongoing natural behavior. Built from off-the-shelf components and 3D-printed modules, each unit delivers near-complete panoramic visual input to walking and flying animals and can operate both in the laboratory and off-grid in the field. This design enables controlled perturbations without removing organisms from their ecological or internal state. Using this system, we access behavioral states that are typically transient and difficult to reproduce. Death’s-head hawkmoths, captured mid-migration, were studied in closed-loop flight for over hundred hours while navigating real world terrains and nightskies along their migratory routes. Mormon crickets, collected from active field swarms, were probed in virtual bands to decode the interactions governing collective motion. In weaver ants, scrambling naturalistic scenes disrupts homing, revealing how navigation depends on structured visual memory and how responses differ between familiar and unfamiliar environments. Across these systems, behavior is not induced but encountered, stabilized, and gently perturbed without breaking its continuity. This allows decision-making to be studied as it is expressed, shaped by both internal state and environmental structure. By uniting ecological realism with experimental precision, MATREX offers a path toward understanding how cognition operates in the real world and how individual decisions scale into collective behavior.

The study of pursuit–evasion (PE) problems has attracted growing interest from researchers in biology, systems and control, and robotics over the past few decades. A typical PE game involves two antagonistic agents: a pursuer and an evader. The pursuer seeks to capture the evader in minimum time, while the evader attempts to avoid capture or delay it for as long as possible. Such problems are commonly formulated as differential games and analyzed using a variety of mathematical tools. This talk introduces a geometric framework for studying pursuit–evasion problems. We first construct isochrones for damped double-integrator systems and use them to derive saddle-point strategies. We then incorporate the isochrone-induced capture time into the reward design of a reinforcement learning algorithm, enabling the learned policies to generalize to more complex scenarios. Finally, using empirical data, we explore how isochrones can be characterized for real animals. These results demonstrate that isochrones offer a novel perspective for understanding and addressing both artificial and natural pursuit–evasion interactions.