Emergence of Self-Organized Amoeboid Movement in a Multi-Agent Approximation of Physarum polycephalum

The giant single-celled slime mould Physarum polycephalum exhibits complex morphological adaptation and amoeboid movement as it forages for food and may be seen as a minimal example of complex robotic behaviour. Swarm computation has previously been used to explore how spatiotemporal complexity can emerge from, and be distributed within, simple component parts and their interactions. Using a particle based swarm approach we explore the question of how to generate collective amoeboid movement from simple non-oscillatory component parts in a model of P. polycephalum. The model collective behaves as a cohesive and deformable virtual material, approximating the local coupling within the plasmodium matrix. The collective generates de-novo and complex oscillatory patterns from simple local interactions. The origin of this motor behaviour is distributed within the collective rendering is morphologically adaptive, amenable to external influence, and robust to simulated environmental insult. We show how to gain external influence over the collective movement by simulated chemo-attraction (pulling towards nutrient stimuli) and simulated light irradiation hazards (pushing from stimuli). The amorphous and distributed properties of the collective are demonstrated by cleaving it into two independent entities and fusing two separate entities to form a single device, thus enabling it to traverse narrow, separate or tortuous paths. We conclude by summarising the contribution of the model to swarm based robotics and soft-bodied modular robotics and discuss the future potential of such material approaches to the field.

💡 Research Summary

The paper presents a particle‑based swarm model that reproduces the amoeboid locomotion of the slime mould Physarum polycephalum using only simple, non‑oscillatory agents. Each agent follows two local rules: (1) chemotaxis toward a virtual nutrient concentration field and (2) a compression‑expansion response based on the distance to neighboring agents. Although individual particles do not possess intrinsic oscillations, the nonlinear coupling of these rules generates spontaneous density waves that propagate through the collective, effectively turning the whole swarm into a deformable, self‑propelled “virtual material.”

Simulation experiments demonstrate that the emergent collective exhibits several hallmark behaviours of the biological organism. First, the swarm produces de‑novo oscillatory patterns that drive coordinated shape changes and forward motion without any central controller. Second, external cues can steer the swarm: simulated chemo‑attraction pulls the material toward nutrient sources, while simulated light irradiation acts as a repellent, causing the swarm to avoid illuminated zones. These responses arise purely from the agents’ local rule set, confirming that global navigation can be achieved through distributed sensing and actuation.

A key contribution of the work is the exploration of modularity and robustness. The authors split a single swarm into two independent sub‑swarms, each of which continues to generate its own oscillatory dynamics and can navigate narrow or tortuous passages individually. When the sub‑swarms are brought back into contact, their internal wave patterns synchronize, and the two bodies fuse into a single coherent entity. Conversely, two separate swarms placed side by side spontaneously merge, illustrating self‑assembly capabilities. The model also tolerates simulated environmental insults such as particle loss; the remaining agents reorganize to preserve overall motion, highlighting intrinsic fault‑tolerance.

From a robotics perspective, the study offers a proof‑of‑concept for soft, amorphous robots that rely on material‑level computation rather than conventional hardware actuators and centralized control. By encoding sensing (chemotaxis, phototaxis) and mechanical response (compression/expansion) directly into the particle interactions, the system achieves adaptive morphology, distributed motor generation, and environmental responsiveness with minimal algorithmic complexity. This aligns with emerging trends in swarm robotics, soft robotics, and modular robotics, where scalability, resilience, and the ability to navigate unstructured environments are paramount.

The authors conclude by suggesting future directions: integrating more realistic physico‑chemical processes (e.g., cytoplasmic streaming), scaling the model to three dimensions, and coupling the swarm with physical soft‑matter substrates to create hybrid bio‑inspired robotic platforms. Overall, the paper advances our understanding of how simple local interactions can give rise to complex, robotically useful behaviours, and it opens pathways for designing next‑generation soft-bodied, self‑organizing robotic systems.