Engineering Swarms Across Scales: From Bird Murmurations to Biomedical and Urban Applications

This talk surveys swarm robotics, showing how simple local rules can produce complex collective behaviors inspired by birds, ants, bees, and cells. It covers bio-inspired controllers, automatic exploration with artificial evolution, reaction-diffusion robot tissues, and scalable hardware from coin-sized robots to distributed systems for cloaking room logistics. Real-world applications span wildfire management, infrastructure inspection, maritime and deep-sea tasks, disaster response, and nanoscale biomedical swarms. The speaker also discusses trust, safety, and verification frameworks for deploying heterogeneous swarms in cities, space, and medicine, highlighting public engagement and a roadmap for the next decades.

Introduction to Swarm Engineering

The speaker introduces swarm robotics as the study of how large numbers of simple, autonomous agents can exhibit collective behaviors that emerge from local interactions with each other and the environment. Drawing inspiration from murmuration, ant trails, bee foraging, and tissue growth, the talk emphasizes key properties such as scalability, robustness to individual failures, and the emergence of complex patterns from simple rules.

Design Tools: Bioinspiration and Automatic Exploration

Two main tools guide swarm design: bioinspiration and automatic exploration. Bioinspiration translates rules observed in biology to artificial agents. The speaker recounts a 15-year-old project implementing flocking rules on flying robots, yielding predictable circular swarms. Simple local rules include attraction to nearby agents, repulsion to avoid collisions, alignment with neighbors, and migration toward a target point. In addition to aerial swarms, simpler coin-sized robots show how rules can create bee-like trails or collective decision making. A modern extension uses reaction-diffusion principles, inspired by Turing, to generate spots and stripes in robotic tissue, with robots acting as cells that exchange two simulated chemicals and diffuse information to produce robust, self-healing patterns.

Automatic Design: Artificial Evolution and Interpretable Brains

The talk then shifts to automatic exploration of swarm programs using artificial evolution. Starting with random programs tested in simulations, the best performers are recombined (crossover) and mutated to produce improved generations. In real deployments, onboard GPUs enable real-time evolution and adaptation, such as pushing a Frisbee in an arena. Understanding the resulting policies is made possible by behavior trees, which provide human-readable representations of the swarm brain. A separate method uses video data to automatically extract the governing rules of a swarm through evolution, enabling interpretable insights into natural swarms and artificial implementations.

Hardware and Real-World Scales

The talk demonstrates hardware at multiple scales, from 1,000 coin-sized robots capable of sensing neighbors within 10 cm to more capable DOTS robots designed for cloakroom logistics. The central idea is to move beyond lab conditions toward real-world environments where central control is impractical; local communication and distributed spatial awareness allow scalable control with resilience to failures. The speaker emphasizes the value of self-organization as a design principle for robust, scalable systems that can operate in field settings without a central planner.

From Lab to Life: Applications and Implications

The presentation surveys macro-scale applications such as search and rescue, infrastructure inspection, ocean cleanup, agriculture, and dynamic city-scale displays. It then transitions to nanoscale swarms for medical applications, where nanoparticles are programmed by design parameters (size, shape, material, surface chemistry, loading) to control distribution and efficacy within tumors. The speaker notes counterintuitive effects where the collective behavior of many particles differs markedly from single-particle expectations, motivating the use of automatic design to optimize delivery. The dome-based Lumidome illustrates microswarm control with light, enabling communication, environmental patterning, and rate-based manipulation of algae and wound healing in biomedical settings.

Human-Swarm Interaction and Trust

Crucial questions center on how humans will interact with swarms in daily life and critical missions. The talk discusses trust, safety, and governance, highlighting a framework for specification, verification, and validation of emergent swarm properties. A practical example demonstrates how to prevent a swarm from blocking fire exits through a tiered testing pipeline: specification, formal verification, fast low-fidelity simulation, realistic simulation, and arena testing. The talk also presents strategies for fault detection and mitigation, showing how local sensing and distributed reasoning help a swarm adapt to faults without central control.

Future Roadmap and Societal Impact

A recurring theme is the scale-driven approach to real-world deployment. The roadmap envisions agriculture, infrastructure inspection, maritime and deep-sea tasks, entertainment, city-scale collaboration with humans, space exploration, and even medical micro- and nano-scale swarms. The speaker contends that swarms will not be monolithic but heterogeneous, with shared goals but diverse hardware and software, necessitating governance that protects safety, privacy, and societal values. Finally, the talk emphasizes the interdisciplinary nature of swarm robotics, drawing from biology, machine learning, computer science, social science, and engineering to build trusted, scalable, and beneficial systems.