Semantic analysis of behavior in a DNA-functionalized molecular swarm

Tom Bachard; Gong Yiming; Ibuki Kawamata; Akira Kakugo; Nathanael Aubert-Kato

cs.ROApr 7, 2026

Semantic analysis of behavior in a DNA-functionalized molecular swarm

Authors: Tom Bachard, Gong Yiming, Ibuki Kawamata, Akira Kakugo, Nathanael Aubert-Kato

Organizations: Ochanomizu University, Tokyo, Japan · University of Kyoto, Kyoto, Japan

Abstract

In this paper, we propose applying semantic embedding to learn the range of behaviors exhibited by molecular swarms, thereby providing a richer set of features to optimize such systems. Specifically, we consider a standard molecular swarm where the individuals are cytoskeletal filaments (called microtubules) propelled by surface-adhered kinesin motors, with the addition of DNA functionalization for further control. We extend a microtubule model with that additional interaction and show that the extracted semantic atoms from simulation results match the expected behaviors. Moreover, the decomposition of each frame in the simulations accurately describes the expected impact of the external control values. Those results provide relevant leads towards the explainability of simulated experiments, making them more reliable for designing and optimizing in-vitro systems.

Figures & tables

Fig. 1: Molecular swarm. (a) At the local scale, microtubules (MTs) are carried over a bed of kinesin grafted to a glass surface. Upon collision, MTs will change their orientation. While such interaction is weak in regular MTs, we use DNA functionalization to enhance the swarming effect: complementary DNA strands attached to the surface of the MTs form duplexes, providing additional support to align the strands. Finally, those duplexes can be destabilized by an external output (here, temperature), reversing the process. (b) At the global scale, depending on the stability of DNA duplexes, we observe three types of behaviors: disorder, when duplexes are unstable; partial swarming, where groups can form with MTs continuously joining and leaving; swarming, when large groups recruit many MTs, with nearly no disassembly.

Fig. 2: Forces applied to the microtubules. Forces based on a Lennard-Jones potential $F^{lj}$ corresponds to microtubule-microtubule interactions, while forces based on the DNA potential $F^{dna}$ corresponds to the strength of grafted DNA duplexing interactions. When the temperature increases, that interaction becomes weaker, according to the stability of the structure.

Panel of Figure 3 — Fig. 3: Selected frames from the C-GLASS simulations with the given parameters. The color represents the angle of the microtubules relative to an arbitrary fixed reference. (Top to bottom, left to right) Random initialization common to all the simulations. Middle of the simulation for a temperature of $200$ K. End of a simulation for a temperature of $200$ K. End of the simulation for a temperature of $275$ K. Middle of the simulation for a temperature of $400$ K. End of a simulation for a temperature of $400$ K.

Fig. 4: Learning a semantic dictionary from a given image collection

\mathcal{X}

. Pipeline from [ 4 ] .

Fig. 5: Sample images generated from the atoms of a semantic dictionary learned on the frame of the simulations. (Top to bottom, left to right) Ordered atoms by their relevancy ( i.e., by their discriminant capacity) in the dictionary.

Panel of Figure 6 — Fig. 6: Box plot activations for some atoms of the dictionary for the different temperatures of the simulations. The remaining atoms activation maps can be found in the appendix. The first $50$ frames of each simulation are not accounted in the computation as they do not represent typical behavior due to the non-physical random initialization.

Panel of Figure 7 — Fig. 7: Semantic decomposition of frames in the dictionary. (Top to bottom) Initial random frame common to all the simulations. Final frame of the simulations for a temperature of respectively $200K$ , $300K$ and $400K$ .

Panel of Figure 8 — Fig. 8: Temporal evolution of the atoms’ activations for different experiments. (Left to right) Temperature respectively set to $200K$ , $300K$ and $400K$ .

Fig. 9: External temperature estimation using the semantic activation for each frame.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 10 — Fig. 10: Box plot activations for all the $12$ atoms of the dictionary for the different temperatures of the simulations. The $50$ first frames of the simulations are not accounted as they only describe the initialization of the simulations rather than the studied permanent regime.

Panel of Figure 11 — Fig. 11: Temporal evolution of the activations of the atoms for all the different temperatures set-ups. From top to bottom, left to right, the temperature is $200+i\times 25K,\ i\in[0,...8]$ .

Emergent Behavior in Multi-Agent Systems

Emergent behavior in multi-agent systems studies how local agent interactions produce collective patterns such as cooperation, norms, and instability.

See all 170 papers

Physics-Informed Modeling and Control of Emergent Behaviors in Robot Swarms
Jun 1, 2026Zixuan Jin, Wenzhuo Zhang, Shuxian Quan +4Multi-Robot Systems Swarm Robotics
School of Computer and Communication Engineering, University of Science and Technology Beijing, 100083, Beijing, China. · Shunde Innovation School, University of Science and Technology Beijing, 528399, Foshan Guangdong, China.
SenseWalk: Agent-Based Semantic Trajectory Simulation Powered by Large Language Models in Zoned Environments
Jul 1, 2026Ziyue Lin, Xinhang Xie, Kangyi Wang +1Pedestrian Behavior Modeling Trajectory Generation
School of Data Science, Fudan University · Shanghai, China
Behavioral Probes for Information Flow in LLM Swarms
Oct 19, 2026Junhui Chen, Yu Zhang, Lantian Li +1Multi-Agent LLM Systems LLM Agent Workflow Optimization
School of Artificial Intelligence, Beijing University of Posts and Telecommunications, China · Center for Speech and Language Technologies, BNRist, Tsinghua University, China

Jun 1, 2026J/K move · Enter open · S save

Physics-Informed Modeling and Control of Emergent Behaviors in Robot Swarms

Zixuan Jin, Wenzhuo Zhang, Shuxian Quan, Zirui Dong+3

School of Computer and Communication Engineering, University of Science and Technology Beijing, 100083, Beijing, China. · Shunde Innovation School, University of Science and Technology Beijing, 528399, Foshan Guangdong, China.

Multi-Robot Systems Swarm Robotics Multi-Agent Control Physics-Informed ML

Open paper AskPDF arXiv

Robot swarms can exhibit coherent collective behaviors through local perception, limited communication and decentralized decision-making, yet modeling and controlling such emergence remains challenging when behaviors unfold over multiple phases. Here we introduce PhySwarm, a physics-informed micro--macro framework that represents multi-stage swarm emergence as physically constrained density-field evolution coupled to executable robot motion. At the macroscopic level, a multi-phase advection--diffusion--reaction model (Macro-ADR) describes phase-dependent swarm-density evolution through directed transport, diffusion-based spatial regulation and behavioral phase transitions. At the microscopic level, an equivalent deterministic motion model (Micro-EDM) realizes these mechanisms through potential-field advection, density-gradient compensation and rate- or event-gated phase switching. A neural-physics controller (NPC) maps local observations and temporal memory to bounded physical parameters, and is trained with a reinforcement learning--PINN objective that combines task rewards with macro-scale density residuals and micro-scale motion-consistency constraints. In several proof-of-concept swarm missions -- including trail-guided foraging, formation-reconfigurable navigation and role-adaptive search and rescue -- we demonstrate that PhySwarm can generate distinct multi-stage emergent behaviors within a unified physics-informed modeling framework. The learned density fields and physical parameters provide interpretable evidence of how advection, diffusion and reaction jointly regulate multi-stage swarm organization. These results establish a physics-informed route for learning, interpreting and controlling emergent behaviors in robot swarms.

Abstract

Emergent Behavior in Multi-Agent Systems

Explore similar work

Physics-Informed Modeling and Control of Emergent Behaviors in Robot Swarms

SenseWalk: Agent-Based Semantic Trajectory Simulation Powered by Large Language Models in Zoned Environments

Behavioral Probes for Information Flow in LLM Swarms