Direction-Aware A* Planning and Numerically Stable Quintic Trajectory Generation for AGV

Abstract

High-precision control of heavy-duty autonomous ground vehicles (AGVs) in cluttered factories requires paths that account for the full vehicle–cargo envelope and numerically reliable reference trajectories across widely varying segment durations. This paper presents a two-stage planning and reference-generation framework that converts occupancy-grid routes into geometrically consistent, dynamically bounded references for high-precision control. The planning front end augments grid search with previous-axis state, conservative envelope collision checking, and action-preserving compression to retain maneuver boundaries while reducing avoidable axis alternations. The trajectory back end generates fixed-duration minimum-jerk quintics in a dimensionless basis, applies precomputed derivative scaling, evaluates derivative-specific continuous-time extrema, and adapts segment durations to configured velocity, acceleration, and jerk limits. Across 160 paired tasks on four occupancy-grid maps, the method preserves mean path length while reducing mean axis switches by 13.2%–69.8% and compressed waypoints by 9.7%–43.1%. Over T ∈ [10−3, 103] s, the normalized system maintains a condition number of 874, while the raw power-basis system reaches 2.41×1016. In 300 general boundary cases, derivative-specific extrema recover all continuous polynomial peaks; the engineering candidate set underestimates velocity peaks in 95.0% of cases, with a maximum error of 46.96%. Together, these tests evaluate the geometric and numerical prerequisites of high-precision control while leaving feedback tracking error outside the present scope. These results establish the framework as a planning and reference-generation basis for subsequent high-precision tracking control. Its output is an open-loop reference; closed-loop tracking accuracy remains to be evaluated with the tracking controller.