Optimization-based motion primitive automata for autonomous driving

TL;DR

This work presents an optimization-driven framework for autonomous driving that uses a finite set of motion primitives organized as a universal automaton (ua). By solving multiobjective optimal control problems, the authors generate Pareto-optimal maneuvers that connect fixed-control trims within a symmetry-aware kinematic model, and they enable reconfiguration of the automaton to prioritize different criteria at planning time. A receding-horizon graph-search (rhgs) algorithm operates on subgraphs tailored to the current objective, balancing travel time, effort, and comfort. Evaluations in simulation and on a 1:18-scale lab platform demonstrate that the choice and optimization of primitives significantly influence real-time performance, trajectory quality, and computation, highlighting the need for dynamic tradeoffs in planning. The approach offers a flexible, scalable method for multiobjective trajectory planning in networked autonomous vehicles, with potential for real-time reconfiguration based on changing priorities.

Abstract

Trajectory planning for autonomous cars can be addressed by primitive-based methods, which encode nonlinear dynamical system behavior into automata. In this paper, we focus on optimal trajectory planning. Since, typically, multiple criteria have to be taken into account, multiobjective optimization problems have to be solved. For the resulting Pareto-optimal motion primitives, we introduce a universal automaton, which can be reduced or reconfigured according to prioritized criteria during planning. We evaluate a corresponding multi-vehicle planning scenario with both simulations and laboratory experiments.

Figures (4)

• Figure 1: Example of a Pareto set with respective positions, orientations, and inputs for a maneuver going from a trim with zero speed and steering angle to $v=2.3m\per s$ and $\delta = 0.62rad$. It was considered the kst model with $J_1= -\int_0^T (||s_x||_2^2 + ||s_y||_2^2) dt$ and $J_2=\int_0^T ||u||_2^2dt$, for a fixed duration $T=0.2s$, and the parameters used in the evaluation.
• Figure 2: The universal automaton used in the evaluation for the trims given in \ref{['tab:trims']}. Blue maneuvers were computed for $J_1$, red ones were computed for $J_2$, and green ones, for $J_3$. Bidirectional arrows represent two maneuvers, one in each direction.
• Figure 3: Test scenario for three cars.
• Figure 4: Simulation of the red vehicle.