[Paper Review] Expected Coverage of Random Walk Mobility Algorithm

Moritz Kohls, Tanja Hernandez|arXiv (Cornell University)|Nov 9, 2016

Optimization and Search Problems2 references3 citations

TL;DR

This paper presents an analytical solution for the expected coverage of the symmetric random walk mobility algorithm in UAV swarms using Markov chains with absorbing states to model coverage over time. It demonstrates that ignoring the dependence between coverage events leads to significant inaccuracies, validating the correct model through simulations and showing that MCMC simulations provide a reliable alternative for large-scale systems.

ABSTRACT

Unmanned aerial vehicles (UAVs) have been increasingly used for exploring areas. Many mobility algorithms were designed to achieve a fast coverage of a given area. We focus on analysing the expected coverage of the symmetric random walk mobility algorithm with independent mobility. Therefore we proof the dependence of certain events and develop Markov models, in order to provide an analytical solution for the expected coverage. The analytic solution is afterwards compared to those of another work and to simulation results.

Motivation & Objective

• To derive an accurate analytical solution for expected coverage of symmetric random walk mobility in UAV swarms.
• To identify and correct the error in prior work that assumed independence of coverage events.
• To validate the corrected analytical model using MCMC simulations.
• To compare the performance of the corrected model with previous approximations and simulation results.

Proposed method

• Model the UAV mobility as a discrete-time, homogeneous Markov chain on a 2D grid with states representing grid positions.
• Introduce absorbing states to model first-visit events, enabling calculation of first-passage time probabilities.
• Use transition matrices P_r with a 1 at diagonal (r,r) to represent absorption at state r, allowing computation of P(X_n = r) via matrix exponentiation.
• Derive expected coverage as the average of first-visit probabilities across all states: E[coverage] = (1/|E|) * Σ_r (π₀ P_r^n)_r.
• Extend the model to multi-UAV independent mobility using the formula P(C_n,r^multi=1) = 1 - (1 - P(C_n,r=1))^|E|.
• Validate results via MCMC simulations to confirm the correctness of the analytical model against naive independence assumptions.

Experimental results

Research questions

• RQ1 What is the true analytical expression for expected coverage under the symmetric random walk mobility model, accounting for event dependence?
• RQ2 Why does the prior analytical solution in [1] produce incorrect results, and what is the root cause of the error?
• RQ3 How does the assumption of independent coverage events affect the accuracy of expected coverage estimation?
• RQ4 To what extent do MCMC simulations match the corrected analytical solution?
• RQ5 Can the corrected model be practically applied to large-scale UAV swarms despite computational complexity?

Key findings

• The assumption of independent coverage events in prior work leads to a systematically incorrect analytical solution, not due to numerical error.
• The correct analytical solution requires modeling first-visit times using absorbing Markov chains, where each state r has its own transition matrix P_r with absorption at r.
• The expected coverage after n steps is given by E[coverage] = (1/|E|) * Σ_r (π₀ P_r^n)_r, which accounts for state visit dependence.
• MCMC simulations confirm the correctness of the derived analytical model and show strong agreement with the exact solution.
• For large state spaces, MCMC simulations are preferred over direct analytic computation due to numerical intractability of matrix exponentiation.
• The paper demonstrates that naive independence assumptions in coverage metrics lead to significant performance overestimation, invalidating prior results.