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

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 the single expert model in managing diverse scenarios, and insufficient consideration of environmental interactions. T o 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 SOT A 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. UTONOMOUS driving trajectory planning has evolved over decades, with rule-based methods [1]-[3] providing fundamental safety assurances via predefined logic and heuristics. However, in complex urban settings, three significant limitations become apparent: (1) The manual construction of rules struggles to accommodate dynamic interactions and abrupt changes in road topology, resulting in unaddressed long-tail scenarios; (2) Rigid trajectory generation fails to mimic the adaptive behaviors of human drivers, such as dynamically adjusting following distances; (3) An exponential increase in maintenance costs arises from the "combinatorial explosion" of accumulating rules. Conversely, data-driven approaches, including imitation learning [4]-[6], address edge cases like extreme weather and complex traffic, capturing human-like driving behaviors from expert data. Reinforcement learning [7], [8] enables dynamic optimization through advanced reward mechanisms. These systems offer lower costs and faster iterations compared to rule-based alternatives.