Name: ISRAEL NARDOTO BARBOZA
Publication date: 31/07/2026
Examining board:
Name |
Role |
|---|---|
| ALBERTO FERREIRA DE SOUZA | Presidente |
| ANDRE GEORGHTON CARDOSO PACHECO | Examinador Interno |
| CLAUDINE SANTOS BADUE | Coorientador |
| FRANCISCO DE ASSIS BOLDT | Examinador Externo |
Summary: This work presents NLP-MATPlan (Nonlinear Programming-based Multi-Articulated
Trajectory Planning), a trajectory planning framework developed to enable safe and efficient
autonomous maneuvers for multi-articulated trucks in complex industrial facilities. The
primary objective of this framework is to compute a precise, physically feasible trajectory
from an initial state to a target destination, accounting for the vehicle’s kinematics and
environmental constraints. Furthermore, the framework is designed to be flexible, seamlessly
adapting to a variety of vehicle platforms and multi-articulated compositions. Consequently,
it explicitly addresses key challenges inherent to these vehicles, including underactuation,
nonholonomic constraints, off-tracking, and jackknifing during reverse maneuvers. The
proposed approach decouples global exploration from continuous optimization: a Hybrid
A* search method, that evaluates a comprehensive cost function. This function integrates
a cost-to-come, which penalizes obstacle proximity and complex directional maneuvers,
with a heuristic cost-to-go that leverages both holonomic and nonholonomic estimations
to efficiently guide node selection. To ensure a smooth transition from discrete exploration
to continuous optimization, a direct multiple shooting NLP formulation is systematically
seeded at each iteration by linearly interpolating between the minimum-cost candidate
state extracted from a binary heap and the final goal destination via warm-start. By
modeling the vehicle kinematics through a system of Ordinary Differential Equations
(ODEs), the NLP stage generates a smooth, time-parameterized, full-state trajectory.
Specifically, the solver outputs the complete kinematically compliant configuration of
the vehicle at each time-step, comprising the tractor’s planar pose, longitudinal velocity,
steering angle, and the vector of articulation angles. Safety is guaranteed by incorporating
an efficient circle-based collision-checking strategy to validate the planned trajectory. The
framework is integrated into a modular SAE Level 4 autonomous architecture derived from
the Intelligent Autonomous Robotic Automobile (IARA) platform and extended for full-
scale heavy trucks. Validation encompasses simulations of multi-articulated configurations
and real-world autonomous docking field tests. In these scenarios, maintaining a terminal
lateral error below 0.15 m is a strict physical requirement of the dock to ensure proper
mechanical coupling. The results demonstrate that NLP-MATPlan consistently generates
feasible, collision-free trajectories while successfully compensating for initial lateral errors
of up to 0.35 m. The system reliably executes these reverse docking maneuvers, achieving
a terminal heading deviation under 1° and keeping the semi-trailer’s terminal lateral error
below 0.07 m. Combined with an average CPU processing time under 1 s, the framework
proves its readiness for real-time industrial applications.
