Maneuver Decision-Making with Trajectory Streams Prediction for Autonomous Vehicles

TL;DR

FFStreams++ addresses the challenge of safe, efficient maneuver decision-making for autonomous vehicles by integrating trajectory prediction with task-and-motion planning. It combines maneuver-specific trajectory streams, Frenet/jerk-optimized trajectory generation, and a PDDL2.1 formulation solved by FastForward, augmented with a Query-Centric network (QCNet) that predicts surrounding obstacles' futures. The framework demonstrates higher success rates and more human-like driving behavior than a baseline search-based planner in CommonRoad intersection and highway scenarios, while maintaining safety and passenger comfort through a composite cost that emphasizes jerk minimization and adherence to acceleration/curvature limits. This predictive, streaming-planning approach enables real-time re-planning with sub-200 ms latency, supporting robust operation in dynamic, heterogeneous traffic environments. The work advances autonomous driving by tightly coupling trajectory prediction with integrated decision-making and motion planning in a scalable, interpretable planning framework.

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.

Figures (8)

• Figure 1: A critical unprotected left-turn maneuver at an unsignalized intersection.
• Figure 2: A critical overtaking maneuver in a highway scenario.
• Figure 3: The schematic diagram of the proposed autonomous driving system is presented. It incorporates high-definition map information alongside the ego vehicle's current state and the present and past states of obstacles, as input from the perception module. The decision-making and motion planning framework, utilizing trajectory prediction, formulates the optimal trajectory, providing acceleration data and headings as outputs. These outputs serve as inputs to the control module within the autonomous driving system.
• Figure 4: Scheme of the proposed FFStreams++ framework for integrated decision-making and motion planning with trajectory prediction in autonomous driving within dynamic environments. The system combines trajectory prediction with a target reference line derived from intersection and highway scenarios to generate an initial PDDL problem. This problem is iteratively modified by considering various trajectory streams (Yield, Follow Speed, Right Change, and Left Change Streams) until an optimal plan is achieved. The problem is formulated in PDDL2.1, solved using a FastForward Planner, and the resulting optimal plan is dispatched to guide the vehicle's decision-making and trajectory.
• Figure 5: The RMSE of the prediction model against different prediction horizons.