Efficient and Interaction-Aware Trajectory Planning for Autonomous Vehicles with Particle Swarm Optimization

Song, Lin; Isele, David; Hovakimyan, Naira; Bae, Sangjae · 2024 · Crossref

DOI: 10.1109/iv55156.2024.10588678

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 interaction-aware trajectory planning for autonomous vehicles in dense traffic, specifically focusing on safe and smooth lane-change maneuvers. Existing approaches, such as game-theoretic methods, are often computationally expensive and do not scale well, while reinforcement learning techniques struggle with reliability and interpretability. Although Neural Network Model Predictive Control (NNMPC) integrates data-driven predictions with control mechanisms, it faces an accuracy-efficiency trade-off; sampling-based strategies require generating numerous trajectories, and analytic solutions like the Alternating Direction Method of Multipliers (ADMM) are too slow for real-time applications. To resolve this, the authors propose a novel numerical approach that combines Particle Swarm Optimization (PSO) with polynomial curve fitting and neural network predictions to achieve real-time, dynamically feasible, and smooth trajectories. The proposed method utilizes a nonlinear kinematics bicycle model for vehicle dynamics and a Social Generative Adversarial Network (SGAN) to predict the future positions of surrounding vehicles based on historical observations. The core planning algorithm employs PSO, where each particle represents a sequence of steering angles. These particles are propagated through the vehicle kinematics to generate candidate trajectories. A comprehensive cost function evaluates each candidate, penalizing deviations from a reference trajectory, heading errors, acceleration and jerk (for comfort), and lane-center misalignment. Crucially, safety constraints are integrated by assigning high costs to trajectories that violate inter-vehicle distance metrics, which are calculated using a three-circle vehicle representation and SGAN-predicted positions of other agents. To ensure smoothness, the best feasible trajectory found by PSO is refined using a third-order polynomial curve fit, which generates new reference waypoints for iterative optimization. The authors validated the approach through numerical simulations comparing it against two baselines: an analytic ADMM-NNMPC method and a heuristic Monte-Carlo sampling method. The experiments involved lane-change scenarios in dense traffic with randomized initial positions for surrounding vehicles. The results demonstrated that the proposed PSO-based algorithm successfully refined initially unsafe trajectories into collision-free paths while maintaining optimality comparable to the ADMM baseline. Notably, the unoptimized implementation of the proposed method achieved computation times of less than 100 milliseconds, confirming its capability for real-time execution. The simulations showed that the method effectively balances safety, driving comfort, and trajectory smoothness, successfully navigating tight gaps between vehicles where baseline methods either failed or were computationally prohibitive. The significance of this work lies in its ability to bridge the gap between learning-based interaction prediction and efficient numerical optimization. By integrating SGAN predictions directly into the PSO search process, the algorithm accounts for the interactive behavior of surrounding vehicles without the heavy computational burden of traditional MPC solvers. This approach offers a practical solution for autonomous driving systems requiring real-time decision-making in complex, dense traffic environments, ensuring both safety and passenger comfort through smooth, dynamically feasible trajectory generation.

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.

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.

• situational awareness