[Paper Review] Optimal mass transport and kernel density estimation for state-dependent networked dynamic systems

Mathias Hudoba de Badyn, Utku Eren+2 more|arXiv (Cornell University)|Sep 20, 2018

Mathematical Biology Tumor Growth29 references4 citations

TL;DR

This paper proposes a hybrid control strategy for state-dependent networked multi-agent systems using optimal mass transport (OMT) for feedforward density shaping and state-dependent kernel density estimation (KDE) for local feedback control. The method achieves accurate density tracking by combining a globally optimal velocity field from OMT with decentralized density estimates that respect dynamic network topology, ensuring convergence to the target distribution under sparse, state-dependent communication.

ABSTRACT

State-dependent networked dynamical systems are ones where the interconnections between agents change as a function of the states of the agents. Such systems are highly nonlinear, and a cohesive strategy for their control is lacking in the literature. In this paper, we present two techniques pertaining to the density control of such systems. Agent states are initially distributed according to some density, and a feedback law is designed to move the agents to a target density profile. We use optimal mass transport to design a feedforward control law propelling the agents towards this target density. Kernel density estimation, with constraints imposed by the state-dependent dynamics, is then used to allow each agent to estimate the local density of the agents.

Motivation & Objective

• To address the lack of cohesive control strategies for nonlinear, state-dependent networked dynamic systems where agent interconnections evolve with states.
• To design a physically realizable control law that guides agents from an initial density profile to a desired target density.
• To integrate optimal mass transport for feedforward control with kernel density estimation for local feedback, accounting for dynamic network topology.
• To ensure convergence to the target density using a feedback law derived from locally estimated densities under proximity-based, state-dependent communication constraints.

Proposed method

• Optimal mass transport (OMT) is used to compute a time-varying velocity field that transports the initial agent density ρ₀ to the target density ρ₁, satisfying the continuity equation.
• The OMT solution provides a feedforward control input that defines the desired motion of agents in the mean-field limit.
• A modified kernel density estimation (KDE) framework is developed to estimate local agent density using only neighbor measurements, constrained by state-dependent network topology.
• A quadratic program is formulated to optimize the kernel bandwidth and weights under proximity-based edge switching, ensuring consistent and physically realizable density estimates.
• The feedback control law is derived from the estimated density to correct deviations from the OMT trajectory, enabling robustness to uncertainty.
• A Lyapunov function is constructed to prove asymptotic convergence of the estimated density to the true density, ensuring stability of the closed-loop system.

Experimental results

Research questions

• RQ1 How can optimal mass transport be adapted to design feedforward control laws for state-dependent multi-agent systems with time-varying interconnections?
• RQ2 What is the role of kernel density estimation in enabling decentralized, local feedback control when agent communication depends on state-dependent proximity?
• RQ3 Can a hybrid control strategy combining OMT and state-dependent KDE achieve accurate and stable density tracking in nonlinear, networked systems?
• RQ4 How does the choice of interaction radius h affect the performance and computational load in proximity-based edge switching?
• RQ5 What conditions ensure the consistency and stability of the locally estimated density in the presence of dynamic network topology?

Key findings

• The OMT-based feedforward control law successfully transports agents from an initial 2D Gaussian density to a target ring-shaped density profile, as verified in simulations on a 100×100×100 grid.
• When combined with the feedback control law based on state-dependent KDE, the agents achieve significantly better organization around the target density compared to using only feedforward control.
• The Lyapunov analysis proves that the estimated density converges to the true density in the mean-field limit, ensuring asymptotic stability of the closed-loop system.
• The kernel density estimation is consistent under the condition lim_{N→∞} Nh(N) = ∞, guaranteeing almost sure convergence of the local density estimate to the true density.
• The feedback control law effectively reduces dispersion and improves alignment with the target distribution, as demonstrated by visual comparison in Figures 5 and 6.
• The method is robust to sparse, state-dependent communication, as the KDE framework adapts to changing neighbor sets without requiring global knowledge.