Maneuver Decision-Making with Trajectory Streams Prediction for Autonomous Vehicles

Mais Jamal, Aleksandr PanovarXiv 2024 · DOI · arXiv

Abstract

Autonomous driving in dynamic environments requires integrating high-level symbolic reasoning with low-level continuous motion planning. This paper presents FFStreams++, a framework that bridges this gap using "Trajectory Streams." By combining Planning Domain Definition Language (PDDL) with deep-learning-based trajectory prediction (QCNet), the system enables autonomous vehicles to navigate complex scenarios like unprotected left turns and high-speed highway overtaking. Experimental results in the CommonRoad simulation environment demonstrate significant improvements in success rates and driving smoothness compared to traditional search-based planners.

Introduction

The challenge of autonomous driving lies in the seamless integration of two distinct cognitive processes: high-level strategic decision-making (e.g., determining right-of-way) and low-level physical execution (e.g., calculating steering angles). Traditional systems often struggle with this integration, either oversimplifying the behavior of other agents or lacking the formal logic required for multi-step strategic planning. This research addresses the problem of Integrated Task and Motion Planning (ITMP) in highly dynamic, high-risk environments.

Visualization of an unprotected left turn scenario at a three-way intersection.

Methodology: The FFStreams++ Framework

The core contribution of this work is the FFStreams++ architecture, which utilizes three primary components to bridge logic and physics:

  1. PDDL (Planning Domain Definition Language): This formal language models the driving logic, defining rules and possible actions—such as lane keeping, yielding, or overtaking—as a structured search problem.
  2. Streams: These conditional functions serve as the interface between discrete logic and the continuous physical world. They generate maneuver-specific trajectory candidates based on environmental state, allowing the PDDL planner to evaluate the logical validity of physical paths.
  3. QCNet (Query-Centric Network): A high-fidelity prediction module trained on real-world data (Argoverse 2). QCNet predicts multiple potential future paths for surrounding obstacles, providing the planner with a probabilistic "crystal ball" of human driver behavior.
The FFStreams++ framework architecture integrating trajectory prediction and PDDL-based planning.

The system employs a Fast-Forward (FF) heuristic search to navigate the planning space. It iteratively samples and optimizes trajectories for passenger comfort (minimizing "jerk") while ensuring safety through collision checks against predicted obstacle paths.

Experimental Evaluation

The framework was evaluated using the CommonRoad simulation framework across two primary scenarios:

  • Intersection Navigation (DEU_Nuremberg-39): Testing unprotected left turns at a complex three-way intersection.
  • Highway Maneuvers (USA_US101-22): Testing high-speed overtaking and lane changes.

Key Results

FFStreams++ was compared against a standard Search-Based Planning Library (SBPL) baseline. The results indicate a substantial leap in performance:

  • Success Rates: In unprotected left turns, FFStreams++ achieved an 84% success rate, nearly doubling the baseline's 53%. In highway overtaking, the success rate reached 92%, compared to just 34% for the baseline.
  • Driving Quality: Using the Occupant’s Preference Metric (OPM), the framework's behavior was categorized as "Normal," whereas the baseline often exhibited "Aggressive" behavior due to less refined motion primitives.

Discussion and Future Work

A key insight of this research is the value of multi-modal prediction. By accounting for the two most likely trajectories of surrounding vehicles, the planner manages the inherent uncertainty of human behavior. Furthermore, the iterative refinement of planning levels ensures that the system finds optimal paths without exhaustive initial computation.

Despite these gains, limitations remain. The current prediction model lacks explicit knowledge of traffic rules, occasionally leading to trajectories that violate road boundaries. Additionally, "decision hesitancy" can occur in highly complex intersections when obstacle predictions fluctuate. Future research will focus on incorporating broader semantic information, such as traffic lights and pedestrian crosswalks, to further enhance urban navigation capabilities.

Summary updated

Abstract

Decision-making, motion planning, and trajectory prediction are crucial in autonomous driving systems. By accurately forecasting the movements of other road users, the decision-making capabilities of the autonomous system can be enhanced, making it more effective in responding to dynamic and unpredictable environments and more adaptive to diverse road scenarios. This paper presents the FFStreams++ approach for decision-making and motion planning of different maneuvers, including unprotected left turn, overtaking, and keep-lane. FFStreams++ is a combination of sampling-based and search-based approaches, where iteratively new sampled trajectories for different maneuvers are generated and optimized, and afterward, a heuristic search planner is called, searching for an optimal plan. We model the autonomous diving system in the Planning Domain Definition Language (PDDL) and search for the optimal plan using a heuristic Fast-Forward planner. In this approach, the initial state of the problem is modified iteratively through streams, which will generate maneuver-specific trajectory candidates, increasing the iterating level until an optimal plan is found. FFStreams++ integrates a query-connected network model for predicting possible future trajectories for each surrounding obstacle along with their probabilities. The proposed approach was tested on the CommonRoad simulation framework. We use a collection of randomly generated driving scenarios for overtaking and unprotected left turns at intersections to evaluate the FFStreams++ planner. The test results confirmed that the proposed approach can effectively execute various maneuvers to ensure safety and reduce the risk of collisions with nearby traffic agents.
BibTeX
@misc{jamal2024maneuver,
  author = {Mais Jamal and Aleksandr Panov},
  title = {{Maneuver Decision-Making with Trajectory Streams Prediction for Autonomous Vehicles}},
  year = {2024},
  eprint = {2409.10165},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2409.10165},
}
Figure