Hierarchical autonomous-driving stacks separate graph-based global routing from local path generation. On compute-limited prototype platforms, converting sparse node sequences into drivable paths is challenging because field-based methods capture road morphology but discard graph semantics, whereas topological routes are discontinuous at joints and not directly trackable. This paper presents an end-to-end refinement pipeline that converts the node sequence of a hierarchical multi-goal A⋆ planner into a continuous, curvature-bounded, lane-centred pose sequence with local overtake capability. An offline Gaussian artificial-potential-field distance map provides a C∞ morphology prior, while its dashed-free variant acts as an overtake safety gate. Cubic B-splines in lane interiors are combined with Hermite–B´ezier arcs at junctions, and a normalprojected elastic band re-centres the path while preserving pose-to-node correspondence. Obstacles on crossable boundaries trigger a sigmoid-in / hold / sigmoid-out cubic-B´ezier detour confined to one inter-checkpoint segment. Experiments on a 1:10- scale vehicle show improved curvature feasibility and boundary clearance, with online complexity bounded for execution on embedded hardware.
Path Refinement from Graph Routes Using Potential Fields and Elastic Bands / Bonci, A., Brunella, F., Colletta, M., Di Biase, A., Libofsha, A., Prist, M.. - ELETTRONICO. - (2026), pp. 1-8. (31th IEEE International Conference on Emerging Technologies and Factory Automation, ETFA 2026 Wasteras, Sweden september 2026) [10.1109/etfa68377.2026.11724837].
Path Refinement from Graph Routes Using Potential Fields and Elastic Bands
Bonci, Andrea
Primo
;Brunella, FedericoSecondo
;Colletta, Matteo;Di Biase, Alessandro;Libofsha, AngjeloPenultimo
;Prist, MariorosarioUltimo
2026-01-01
Abstract
Hierarchical autonomous-driving stacks separate graph-based global routing from local path generation. On compute-limited prototype platforms, converting sparse node sequences into drivable paths is challenging because field-based methods capture road morphology but discard graph semantics, whereas topological routes are discontinuous at joints and not directly trackable. This paper presents an end-to-end refinement pipeline that converts the node sequence of a hierarchical multi-goal A⋆ planner into a continuous, curvature-bounded, lane-centred pose sequence with local overtake capability. An offline Gaussian artificial-potential-field distance map provides a C∞ morphology prior, while its dashed-free variant acts as an overtake safety gate. Cubic B-splines in lane interiors are combined with Hermite–B´ezier arcs at junctions, and a normalprojected elastic band re-centres the path while preserving pose-to-node correspondence. Obstacles on crossable boundaries trigger a sigmoid-in / hold / sigmoid-out cubic-B´ezier detour confined to one inter-checkpoint segment. Experiments on a 1:10- scale vehicle show improved curvature feasibility and boundary clearance, with online complexity bounded for execution on embedded hardware.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


