Hierarchical Game-Theoretic Planning for Autonomous Vehicles
DOI: 10.1109/icra.2019.8794007
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the challenge of real-time trajectory planning for autonomous vehicles interacting with human drivers, a problem complicated by the mutual dependence of their actions. Existing approaches often treat human drivers as static obstacles or pure followers, leading to conservative or unsafe behaviors known as the "frozen robot" phenomenon. Conversely, solving dynamic games that capture this mutual influence is computationally prohibitive for real-time application due to the high dimensionality of vehicle dynamics and state spaces. The authors propose a hierarchical game-theoretic planning framework that decomposes the interaction into a long-horizon "strategic" game and a short-horizon "tactical" game, enabling real-time performance while accounting for coupled interactions. The method employs a two-level structure. The high-level strategic planner solves a dynamic nonzero-sum game with a long horizon (approximately 5 seconds) using simplified vehicle dynamics and a full closed-loop feedback information structure. This allows the system to compute the optimal value function, representing the expected outcome of strategic interactions, using dynamic programming. Crucially, this level accommodates non-deterministic human models, such as noisy rationality, rather than assuming perfect rationality. The low-level tactical planner then uses this strategic value as a terminal cost in a receding-horizon optimization (approximately 0.5 seconds) with high-fidelity dynamics. This guides the local trajectory optimization toward globally optimal solutions by implicitly extending the planning horizon. Experiments were conducted in a simulated two-lane highway environment involving an autonomous vehicle and a human-driven car. The authors compared the hierarchical approach against a standard tactical planner that lacks long-horizon strategic reasoning. In scenarios including easy merges, hard merges, and overtaking, the hierarchical planner demonstrated superior performance. For instance, in a hard merge scenario where the autonomous vehicle started behind the human, the tactical planner failed to merge into the left lane, whereas the hierarchical planner successfully overtook and merged. In an overtaking scenario, the tactical planner accelerated but then braked to remain behind the human, while the hierarchical planner executed a lane change to overtake. The strategic value effectively incentivized the autonomous vehicle to pressure the human driver to change lanes or to execute its own lane change based on the predicted human response. The significance of this work lies in its ability to balance computational tractability with the need for accurate interaction modeling. By separating the strategic reasoning from tactical execution, the framework allows autonomous vehicles to plan for longer horizons and reason about mutual influence without the computational burden of solving full-dynamics games in real time. The results indicate that this approach yields richer, safer, and more effective autonomous behavior compared to existing techniques, mitigating the local nature of trajectory optimization and avoiding overly aggressive or conservative driving styles. The framework is also agnostic to the specific low-level optimizer, making it adaptable to various planning schemes.
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | OpenAlex-citations | — | — | 1 | 2026-06-18 |
| archive | success | unpaywall | — | — | 2 | 2026-06-25 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-18 |
| chunk | success | chunk | — | — | 1 | 2026-06-18 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-18 |
| promote | success | — | — | — | 1 | 2026-06-18 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-18 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Theoretical Contribution: computational model