Scaling and Trade-offs in Multi-agent Autonomous Systems

TL;DR

This work performs large-scale agent-based simulations in three canonical scenarios: swarm-on-swarm battle, cooperative area search with attrition, and pursuit of scattering targets, and shows the benefits of embedding an optimal path planning loop within this framework which can qualitatively improve the scaling laws that govern the outcome.

Abstract

Designing autonomous drone swarms is hampered by a vast design space spanning platform, algorithmic, and numerical-strength choices. We perform large-scale agent-based simulations in three canonical scenarios: swarm-on-swarm battle, cooperative area search with attrition, and pursuit of scattering targets. We demonstrate that dimensional-analysis and data-scaling, established techniques in physical sciences, can be leveraged to collapse performance data onto scaling functions that are mathematically simple, yet counterintuitive and therefore difficult to predict a priori. These scaling laws reveal success-failure boundaries, including sharp break points. Additionally, we show how this technique can be used to quantify trade-offs between agent count and platform parameters such as velocity, sensing or weapon range, and attrition rate. Furthermore, we show the benefits of embedding an optimal path planning loop within this framework, which can qualitatively improve the scaling laws that govern the outcome. The methods we demonstrate are highly flexible and would enable rapid, budget-aware sizing and algorithm selection for large autonomous swarms.

Figures (12)

• Figure 1: Four snapshots from an engagement where attackers destroy defenders as well as the HVU. Parameter values are $N_a = N_d = 40$, $\lambda_a = \lambda_d = 1$, and $R_a = R_d = 2 \sim d_r$.
• Figure 2: (a) Sample curves of $P_a$ versus $N_d$ for six combinations of other parameters. (b) $P_a$ versus $N_d/N_{a,{\rm eff}}$ for the curves shown in panel (a), plus many more different combinations of parameter values. The color scheme is strictly based on $R_a/R_d$. Note that $N_{a,{\rm eff}}$ is a different number for every curve.
• Figure 3: Plots of $N_{a,{\rm eff}}/N_a$ versus $\lambda_a/\lambda_d$ for five different combinations of parameter values. These curves demonstrate that $N_{a,{\rm eff}} \propto N_a (\lambda_a/\lambda_d)^\alpha$, with $\alpha \approx 0.6$.
• Figure 4: (a) $A$ versus $R_d/R_a$ for three values of $R_a$, where $N_{a,{\rm eff}} \propto A$ (see text). (b) $A$ versus $R_d/d_r$ for the same data.
• Figure 5: Schematic overview of a cooperative underwater search mission.