DNFOMP: Dynamic Neural Field Optimal Motion Planner for Navigation of Autonomous Robots in Cluttered Environment

• DOI: 10.48550/arxiv.2308.03539
• Publication Date: 2023-01-01
• Authors: Maksim Katerishich; Mikhail Kurenkov; Sausar Karaf; Artem Nenashev; Dzmitry Tsetserukou
• Keywords: Planner

Motion planning in dynamically changing environments is one of the most complex challenges autonomous driving. Safety a crucial requirement, along with driving comfort and speed limits. While classical sampling-based, lattice-based, optimization-based methods can generate smooth short paths, they often do not consider dynamics environment. Some techniques it, but rely on updating environment on-the-go rather than explicitly accounting for dynamics, which suitable self-driving. To address this, we propose novel method based Neural Field Optimal Planner (NFOMP), outperforms state-of-the-art approaches terms normalized curvature number cusps. Our approach embeds previously known moving obstacles into neural field collision model to account We also introduce time profiling trajectory non-linear velocity constraints by adding Lagrange multipliers loss function. applied our solve optimal motion problem an urban using BeamNG.tech simulator. An car drove generated trajectories three city scenarios while sharing road obstacle vehicle. evaluation shows that maximum acceleration passenger experience instantly -7.5 m/s^2 89.6% devoted normal accelerations below 3.5 m/s^2. The style characterized 46.0% 31.4% being light rail transit moderate style, respectively.