Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Noncooperative Game-Theoretic Planning for Autonomous Driving under Imperfect Information: A Quasi-Perfect Equilibrium Approach
Achieving Level-5 autonomy is challenging due to the multi-agent planning problem. Emerging approaches use game-theoretic planning to tackle this problem, but they are limited in that they have equilibrium existence and convergence issues, depend on restrictive assumptions, and are limited in handli...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
AUC Knowledge Fountain
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| Summary: | Achieving Level-5 autonomy is challenging due to the multi-agent planning problem. Emerging approaches use game-theoretic planning to tackle this problem, but they are limited in that they have equilibrium existence and convergence issues, depend on restrictive assumptions, and are limited in handling other agents’ mistakes. This thesis proposes a two-player game-theoretic planning framework that addresses these limitations. It guarantees equilibrium existence and convergence while only assuming perfect recall and generates strategies that are robust to other agents’ future trembles. The framework adopts the quasi-perfect equilibrium as the solution concept that is sequentially rational and admissible. The framework’s path planner generates equilibrium waypoints by formulating the problem in sequence form as a linear complementarity problem and solving it via Lemke’s algorithm. Then, a path manager generates geometrically optimal trajectories respecting the vehicle's minimum turning radius using Dubins path optimization. The framework was validated on both overtaking and head-on scenarios and demonstrates the ability to filter out collision-prone equilibria by imposing robustness to other agents’ future trembles. |
|---|