← Back ICRA 2026

Path Planning for Four-Wheel Steering Forklifts in Constrained Spaces: A State-Embedded Hamiltonian Fast Marching Approach

Julien Pascal, Jean-Marie Mirebeau, Benoit Thuilot and Paul Checchin

PDF

AI summary

Key figure (auto-extracted from paper)

A state-embedded Hamiltonian Fast Marching planner delivers deterministic, globally optimal paths for four-wheel steering forklifts while outperforming existing algorithms in tight industrial spaces.

Path planning Four-wheel steering Hamiltonian Fast Marching Non-holonomic constraints Autonomous forklifts Deterministic planning

Problem

Industrial forklifts require deterministic, optimal path planning that respects complex four-wheel steering kinematics and maintains precise target alignment, yet existing planners lack these guarantees and fail to fully exploit four-wheel steering maneuverability.

Approach

The method extends the Hamiltonian Fast Marching algorithm by integrating four-wheel steering constraints, a multi-circle robot footprint, and a hybrid state system to compute optimal navigation functions in configuration space.

Key results

Generalized Hamiltonian Fast Marching to four-wheel steering kinematics
Introduced a hybrid state system for predictable, human-like path generation
Demonstrated superior clearance maintenance and predictability over RRT variants
Released open-source code and standardized benchmarking framework

Why it matters

Bridges the gap between theoretical motion planning and industrial automation by enabling safe, predictable, and highly maneuverable autonomous forklift operations in constrained environments.

Abstract

This paper proposes a global planner tailored to four-wheel steering (4WS) forklifts operating in constrained en- vironments. It extends the Hamiltonian Fast Marching (HFM) to integrate 4WS non-holonomic constraints, multi-circle ge- ometry, and a hybrid state system. This ensures mathematical guarantees of determinism, stability, and optimality, while fully exploiting 4WS maneuverability, properties that existing toolboxes and generic planners do not provide. Through qual- itative and quantitative evaluations across various scenarios, we benchmark our method against leading motion planning algorithms, including RRT, RRT*, Informed-RRT*, and SST, using a standardized framework. Results demonstrate that our approach consistently outperforms existing solutions, particu- larly in maintaining minimum clearance distances and ensuring path predictability. The source code, along with modeling and implementation details necessary for replicating the results and benchmarking against state-of-the-art methods, is publicly available. https://github.com/PascalJulien/State- HFM

Index terms

Nonholonomic Motion Planning Constrained Motion Planning Motion and Path Planning