TrajMoE: Scene-Adaptive Trajectory Planning with Mixture of Experts and Reinforcement Learning

2025-12-07

AI Summary

To address two key limitations in end-to-end autonomous driving trajectory planning—(1) the lack of scene-adaptive trajectory priors and (2) the absence of policy-driven optimization in trajectory evaluation—this paper proposes a Scene-Adaptive Mixture-of-Experts (MoE) framework. First, a multimodal perception backbone identifies driving scenes and dynamically routes inputs to specialized expert subnetworks, enabling scene-specific trajectory prior modeling. Second, reinforcement learning is integrated to optimize the trajectory scoring policy, overcoming the constraints of single-stage supervised training. The method jointly leverages MoE architecture, multi-source sensor fusion, and policy-driven evaluation. Evaluated on the nuScenes-NavSim benchmark (ICCV), it achieves 51.08 points, ranking third and demonstrating significantly improved planning robustness and safety in complex traffic scenarios.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Current autonomous driving systems often favor end-to-end frameworks, which take sensor inputs like images and learn to map them into trajectory space via neural networks. Previous work has demonstrated that models can achieve better planning performance when provided with a prior distribution of possible trajectories. However, these approaches often overlook two critical aspects: 1) The appropriate trajectory prior can vary significantly across different driving scenarios. 2) Their trajectory evaluation mechanism lacks policy-driven refinement, remaining constrained by the limitations of one-stage supervised training. To address these issues, we explore improvements in two key areas. For problem 1, we employ MoE to apply different trajectory priors tailored to different scenarios. For problem 2, we utilize Reinforcement Learning to fine-tune the trajectory scoring mechanism. Additionally, we integrate models with different perception backbones to enhance perceptual features. Our integrated model achieved a score of 51.08 on the navsim ICCV benchmark, securing third place.

Problem

• Adapt trajectory priors to diverse driving scenarios
• Refine trajectory scoring with reinforcement learning
• Integrate perception backbones to enhance features

Innovation

• Mixture of Experts for scenario-specific trajectory priors
• Reinforcement Learning fine-tunes trajectory scoring mechanism
• Integrates multiple perception backbones to enhance features