SwarmCoDe: A Scalable Co-Design Framework for Heterogeneous Robot Swarms via Dynamic Speciation

Andrew Wilhelm; Josie Hughes

cs.ROMar 27, 2026

SwarmCoDe: A Scalable Co-Design Framework for Heterogeneous Robot Swarms via Dynamic Speciation

Authors: Andrew Wilhelm, Josie Hughes

Organizations: Department of Mechanical Engineering, École Polytechnique Fédérale de Lausanne

Abstract

Robot swarms offer inherent robustness and the capacity to execute complex, collaborative tasks surpassing the capabilities of single-agent systems. Co-designing these systems is critical, as marginal improvements in individual performance or unit cost compound significantly at scale. However, under traditional frameworks, this scale renders co-design intractable due to exponentially large, non-intuitive design spaces. To address this, we propose SwarmCoDe, a novel Collaborative Co-Evolutionary Algorithm (CCEA) that utilizes dynamic speciation to automatically scale swarm heterogeneity to match task complexity. Inspired by biological signaling mechanisms for inter-species cooperation, the algorithm uses evolved genetic tags and a selectivity gene to facilitate the emergent identification of symbiotically beneficial partners without predefined species boundaries. Additionally, an evolved dominance gene dictates the relative swarm composition, decoupling the physical swarm size from the evolutionary population. We apply SwarmCoDe to simultaneously optimize task planning and hardware morphology under fabrication budgets, successfully evolving specialized swarms of up to 200 agents -- four times the size of the evolutionary population. This framework provides a scalable, computationally viable pathway for the holistic co-design of large-scale, heterogeneous robot swarms.

Figures & tables

Fig. 1: Overview of the SwarmCoDe algorithm that dynamically determines species for the co-design of heterogeneous robot swarms. Genetic tags and a selectivity gene are used to determine partners for evaluations to complete the collaborative task. A relative dominance gene determines the swarm composition, enabling our technique to scale to swarm sizes of 200 robots, four times the number of individuals in the evolutionary population.

Component	Type	Domain	Phenotypic Expression
Tag	Binary	$\{0,1\}^{L}$	Genetic distance for partner selection
Selectivity	Float	$[0,1]$	Similarity threshold for partners
Dominance	Float	$[0,1]$	Swarm sampling probability
BT Opcodes	Int[]	$[0,13]$	Behavior tree encoding
Radius	Float	$[R_{min},R_{max}]$	Size and carrying capacity
HW Tiers	Categ.	$\{1,2,3\}$	Chassis, battery, and motor tiers

TABLE I: Genome structure and parameter bounds

Component	Symbol	Distance Metric ( $d_{x}$ )
Tag	$w_{tag}$	$(H/L)^{\gamma}$
Hardware	$w_{hw}$	Normalized Euclidean distance
Control (BT)	$w_{BT}$	Fraction of differing opcodes
Tool	$w_{tool}$	Binary difference ( $0$ or $1$ )
Size	$w_{size}$	Normalized $\Delta$ radius

TABLE II: Distance metrics for computing genetic compatibility

Fig. 2: Evaluation pipeline for calculating marginal contribution. Swarms are stochastically assembled using the evolved dominance gene to determine swarm composition ratios. By comparing the performance of a swarm containing the focal individual against a baseline swarm where that individual is replaced by partner elites in an identical environment, the algorithm determines the true fitness of the focal individual.

Opcode	Name
0	SEQ (Sequence)
1	SEL (Selector)
2	COND_HAS_PACKAGE
3	COND_NEAR_PACKAGE
4	COND_NEAR_BASE
5	COND_AM_I_STUCK

TABLE III: Opcodes used for the Behavior Tree representation

Category	Symbol	Value
Distance Weights	$w_{tag},w_{hw},w_{BT}$	$1.0,0.5,0.3$
	$w_{tool},w_{size}$	$0.35,0.7$
Fitness Weights	$W_{delivery},W_{collab\_bonus}$	$100.0,50.0$
	$W_{pickup},W_{energy}$	$1.0,0.03$
	$W_{proximity},W_{closeness}$	$1.0,30.0$
Constants	Speciation Threshold $\delta$	$0.4$

TABLE IV: Hyperparameters and empirical constants used in evaluation

Fig. 3: The simulation environment with 20 agents and 16 individual packages. Robots are tasked with retrieving packages and returning them to the base. Robots with pincher end effectors (green and purple) can lift square packages while robots with suction end effectors (red and blue) can lift circle packages. Darker packages are heavier, and located closer to the base; lighter packages are farther away.

Panel of Figure 8 — Fig. 4: Emerging species and their morphological niches for increasingly complex scenarios. The SwarmCoDe algorithm adapts to increasingly complex tasks that require one, two, and four unique species to efficiently complete.

Panel of Figure 9 — Fig. 5: Evolutionary dynamics and fitness for the “Two Package Types with Distance-Based Weights” scenario. (Left) The species composition of the evolutionary population over time, where different colors represent distinct species. Early on, many species compete to remain in the population, with many going extinct within the first 20 generations. (Center) The species composition of the best team. As the species adapt and evolve, the best team composition changes, with some species remaining in the best team for several generations while others are never included. (Right) The fitness of the best team per generation. The fitness evolves towards more optimal values over time, but fluctuates as different species emerge and go extinct in the population.

Budget	Best Fitness	Team Cost	Total Deliveries	Individ. Package Deliveries	Collab. Package Deliveries	Avg. Energy Used	# of Species
2500	125.2	2500	4	4	0	53.8	1
3000	154.1	3000	10	8	2	41.0	2
4000	182.9	3800	10	10	0	67.6	2
5000	211.6	5100	11	10	1	36.4	3
6000	224.7	6110	12	11	1	46.9	3
8000	244.4	7880	11	9	2	42.4	4

TABLE V: Evolutionary results under different budget constraints

Fig. 6: The fitness of the best team per generation for the “Return on Investment” scenario. The fitness of the best team evolves towards more optimal values over time, but fluctuates as different species emerge and go extinct in the general population.

Fig. 7: Emerging species and their morphological niches for the “Return on Investment” scenario.

Fig. 8: The simulation environment with 200 agents, 80 individual packages, and 32 collaborative packages. Robots are tasked with retrieving packages and returning them to the base. Robots must work together to carry certain package types. Darker packages are heavier, and located closer to the base; lighter packages are farther away.

Panel of Figure 14 — Fig. 9: Speciation and fitness results for the 200 agent robot swarm with 50 individuals in the evolutionary pool. Using the relative dominance gene, the SwarmCoDe algorithm is able to scale to a swarm size that is four times the size of its evolutionary pool.

Swarm Robotics

Swarm robotics studies how groups of robots coordinate through local sensing, communication, and interaction to accomplish shared tasks.

See all 99 papers

Task-Driven Co-Design of Heterogeneous Multi-Robot Systems
Apr 23, 2026Maximilian Stralz, Meshal Alharbi, Yujun Huang +1Multi-Robot Systems
Laboratory for Information & Decision Systems, Massachusetts Institute of Technology, Cambridge, MA 02139
Swarm Skills: A Portable, Self-Evolving Multi-Agent System Specification for Coordination Engineering
May 11, 2026Xinyu Zhang, Zhicheng Dou, Deyang Li +10Multi-Agent Coordination Multi-Agent Orchestration
openJiuwen Team & Gaoling School of Artificial Intelligence, Renmin University of China
Event-Triggered Adaptive Consensus for Multi-Robot Task Allocation
Apr 8, 2026Fidel Aznar, Mar Pujol, Álvaro DíezMulti-Robot Task Allocation Swarm Robotics
Department of Computer Science and Artificial Intelligence, University of Alicante, Alicante, Spain

Apr 23, 2026J/K move · Enter open · S save

Task-Driven Co-Design of Heterogeneous Multi-Robot Systems

Maximilian Stralz, Meshal Alharbi, Yujun Huang, Gioele Zardini

Laboratory for Information & Decision Systems, Massachusetts Institute of Technology, Cambridge, MA 02139

Multi-Robot Systems

Open paper AskPDF arXiv

Designing multi-agent robotic systems requires reasoning across tightly coupled decisions spanning heterogeneous domains, including robot design, fleet composition, and planning. Much effort has been devoted to isolated improvements in these domains, whereas system-level co-design considering trade-offs and task requirements remains underexplored. In this work, we present a formal and compositional framework for the task-driven co-design of heterogeneous multi-robot systems. Building on a monotone co-design theory, we introduce general abstractions of robots, fleets, planners, executors, and evaluators as interconnected design problems with well-defined interfaces that are agnostic to both implementations and tasks. This structure enables efficient joint optimization of robot design, fleet composition, and planning under task-specific performance constraints. A series of case studies demonstrates the capabilities of the framework. Various component models can be seamlessly incorporated, including new robot types, task profiles, and probabilistic sensing objectives, while non-obvious design alternatives are systematically uncovered with optimality guarantees. The results highlight the flexibility, scalability, and interpretability of the proposed approach, and illustrate how formal co-design enables principled reasoning about complex heterogeneous multi-robot systems.

Abstract

Swarm Robotics

Explore similar work

Task-Driven Co-Design of Heterogeneous Multi-Robot Systems

Swarm Skills: A Portable, Self-Evolving Multi-Agent System Specification for Coordination Engineering

Event-Triggered Adaptive Consensus for Multi-Robot Task Allocation