UniPlanner: A Unified Motion Planning Framework for Autonomous Vehicle Decision-Making Systems via Multi-Dataset Integration

TL;DR

The paper tackles robustness in autonomous vehicle motion planning under distribution shifts by proposing UniPlanner, a unified framework that integrates cross-dataset experience. It introduces three innovations: Gradient-Free Trajectory Mapper (GFTM) to learn universal history-to-future correlations with gradient isolation, History-Future Trajectory Dictionary Network (HFTDN) to build and retrieve cross-dataset trajectory guidance, and Sparse-to-Dense (S2D) to balance robust learning with dense deployment priors. A three-phase training regime—GFTM pre-training, main network training, and HFTDN fine-tuning—enables safe, cross-dataset knowledge transfer while avoiding shortcut learning. Extensive experiments on nuPlan with Waymo and Lyft data demonstrate improvements over baselines and validate the cross-dataset knowledge aggregation paradigm for motion planning, signaling a scalable pathway for robust AV decision-making across diverse environments.

Abstract

Motion planning is a critical component of autonomous vehicle decision-making systems, directly determining trajectory safety and driving efficiency. While deep learning approaches have advanced planning capabilities, existing methods remain confined to single-dataset training, limiting their robustness in planning. Through systematic analysis, we discover that vehicular trajectory distributions and history-future correlations demonstrate remarkable consistency across different datasets. Based on these findings, we propose UniPlanner, the first planning framework designed for multi-dataset integration in autonomous vehicle decision-making. UniPlanner achieves unified cross-dataset learning through three synergistic innovations. First, the History-Future Trajectory Dictionary Network (HFTDN) aggregates history-future trajectory pairs from multiple datasets, using historical trajectory similarity to retrieve relevant futures and generate cross-dataset planning guidance. Second, the Gradient-Free Trajectory Mapper (GFTM) learns robust history-future correlations from multiple datasets, transforming historical trajectories into universal planning priors. Its gradient-free design ensures the introduction of valuable priors while preventing shortcut learning, making the planning knowledge safely transferable. Third, the Sparse-to-Dense (S2D) paradigm implements adaptive dropout to selectively suppress planning priors during training for robust learning, while enabling full prior utilization during inference to maximize planning performance.

Figures (9)

• Figure 1: Motion type distribution of historical trajectories in Waymo ref19, Lyft ref18, and nuPlan ref17
• Figure 2: Statistical distribution of correlations between historical and future trajectory motion types across Waymo ref19, Lyft ref18, and nuPlan ref17
• Figure 3: Overview of UniPlanner architecture with three core modules (GFTM, HFTDN, and S2D) and the sequential training pipeline. ➀, ➁, and ➂ denote GFTM training, main network training, and HFTDN training phases, respectively
• Figure 4: Architecture of the History-Future Trajectory Dictionary Network
• Figure 5: Comparison of motion type distributions between source datasets and dictionary