Scene-Adaptive Motion Planning with Explicit Mixture of Experts and Interaction-Oriented Optimization

TL;DR

This work tackles urban autonomous driving trajectory planning by introducing EMoE-Planner, which combines an explicit Mixture of Experts (EMoE) with a shared scene router, scene-specific queries, and an interaction-oriented loss to handle multi-modal planning and ego–agent interactions. The method assigns dedicated experts to seven predefined scene types, focuses attention via scene-specific anchors, and optimizes trajectories through an interaction-aware objective that emphasizes near-term safety and efficiency. Experimental results on the NuPlan dataset show that EMoE-Planner outperforms state-of-the-art baselines in closed-loop non-reactive simulations and even surpasses rule-based post-processing in many cases, marking a milestone for pure-learning trajectory planning in complex urban scenarios. The findings indicate strong potential for real-time deployment, improved safety, and reduced reliance on hand-crafted post-processing, while also highlighting areas for future work in highly interactive or adversarial settings and real-vehicle validation.

Abstract

Despite over a decade of development, autonomous driving trajectory planning in complex urban environments continues to encounter significant challenges. These challenges include the difficulty in accommodating the multi-modal nature of trajectories, the limitations of single expert model in managing diverse scenarios, and insufficient consideration of environmental interactions. To address these issues, this paper introduces the EMoE-Planner, which incorporates three innovative approaches. Firstly, the Explicit MoE (Mixture of Experts) dynamically selects specialized experts based on scenario-specific information through a shared scene router. Secondly, the planner utilizes scene-specific queries to provide multi-modal priors, directing the model's focus towards relevant target areas. Lastly, it enhances the prediction model and loss calculation by considering the interactions between the ego vehicle and other agents, thereby significantly boosting planning performance. Comparative experiments were conducted on the Nuplan dataset against the state-of-the-art methods. The simulation results demonstrate that our model consistently outperforms SOTA models across nearly all test scenarios. Our model is the first pure learning model to achieve performance surpassing rule-based algorithms in almost all Nuplan closed-loop simulations.

Figures (13)

• Figure 1: Framework of EMoE-Planner
• Figure 2: Differences between vanilla MoE and EMoE
• Figure 3: Scene-specific anchors
• Figure 4: Weight decay factor
• Figure 5: Closed-loop simulation of roundabout. Subfigures (a) through (d) depict snapshots at 0s, 8s, 12s, and 15s for DiffusionPlanner, subfigures (e) through (h) represent snapshots at corresponding times for Pluto and subfigures(i) through (l) denote snapshots for EMoE-Planner. The ego vehicle is represented by an orange rectangle, while a gray rectangle denotes its ground truth position at the same timestamp. The planned trajectory is shown as a colored solid line, the historical trajectory in purple, and the ground-truth historical trajectory in blue. The candidate planned trajectories are displayed as gray lines.