Journal of Basic and Applied Research International

Path planning for unmanned ground vehicles (UGVs) in uncertain environments requires more than geometric shortest-path optimisation, particularly when narrow-passage traversability depends on imperfect obstacle-position estimates. This paper presents a dependence-aware joint chance-constrained A-Star-based planning framework in which successive passage widths are modelled as correlated random variables. The dependence arises naturally when adjacent passages share an uncertain obstacle boundary: a displacement that enlarges one passage may simultaneously reduce the neighbouring passage. The proposed formulation maps obstacle-boundary uncertainty to passage-width covariance through a linear difference model, evaluates route-level feasibility using a multivariate Gaussian joint probability, and selects the shortest candidate route satisfying a prescribed reliability threshold. The framework is evaluated using two-passage validation, sensitivity analysis, a controlled route-selection testbed, a multi-passage extension, a randomised campaign of 5,000 scenarios, Monte Carlo verification, and runtime benchmarks. In the controlled testbed, the independence approximation accepts the shorter route with a predicted feasibility of 0.901279, whereas the dependence-aware model gives 0.898714 and Monte Carlo gives 0.898332, confirming a threshold-sensitive false-safe acceptance at P_min = 0.90. In the randomised campaign, 58 independence-based false-safe planner decisions were identified at P_min = 0.90; in all 58 cases, the dependence-aware estimate was closer to Monte Carlo, reducing mean absolute error by 84.82% and root-mean-square error by 81.47%. A threshold sweep from 0.85 to 0.99 further showed that false-safe independence-based decisions persisted across all tested reliability levels, while becoming less frequent as the acceptance criterion became more restrictive. Dependence-aware rerouting incurred an average path-length penalty of 18.34% in the 21 cases where a feasible alternative was available. For two correlated passages, the additional probability-evaluation cost was small in absolute terms, whereas higher-dimensional multivariate Gaussian evaluation exhibited a clear scalability limitation. These results indicate that explicitly modelling shared uncertainty improves route-level feasibility estimation and reduces threshold-sensitive false-safe selections in probabilistic graph-based UGV planning.