A Cognitive Hierarchical Framework for Multivehicle Collision-Free Traffic Management of Unregulated Intersections

Abstract

Autonomous vehicles are becoming increasingly popular, requiring efficient and reliable methods for trajectory generation and path planning. A common approach is to convert the original continuous-time optimization problem into a discrete-time parametric optimization problem, but consider only a set of control/action variables, which can limit the feasibility and generality of the generated trajectories. To address these limitations, this thesis proposes an approach to autonomous collision-free vehicle trajectory optimization using model predictive control (MPC) with a continuous range of control inputs. The considered formulation enables the computation of more accurate and dynamically feasible paths for autonomous vehicles. A key challenge in path planning of autonomous vehicles is to account for not only the presence of other vehicles, but also their trajectories. More specifically, the other vehicles must be treated as dynamic obstacles and collision-free trajectories have to be generated. Naturally, one has to take into account and model vehicle-to-vehicle interactions and incorporate such interactions, in terms of state path constraints, as part of the formulation of trajectory optimization problems. We leverage Cognitive Hierarchy Theory, specifically using level-$k$ game theory, to predict the behavior of other vehicles/agents, which affects their trajectories. This framework provides a structured way to anticipate interactions and make informed decisions for designing collision-free trajectories.

For testing the proposed framework, an unregulated crossroad was simulated with two vehicles approaching it at the same time. The conducted experiments were split into two sets of scenarios: with and without the level estimator enabled, which allows one vehicle to change the reasoning depth over the course of simulations. Numerical results demonstrate that the proposed approach is both feasible and effective in capturing realistic vehicle interactions and in generating collision-free trajectories. The level-k game-theoretic modeling enhances the decision-making capabilities of autonomous vehicles, leading to safer and more efficient navigation. These findings highlight the potential of expanding the set of control actions in improving autonomous vehicle trajectory planning.