ARTEMIS: Autoregressive End-To-End Trajectory Planning with Mixture of Experts for Autonomous Driving
Renju Feng, Ning Xi, Duanfeng Chu, Rukang Wang, Zejian Deng, Anzheng Wang, Liping Lu, Jinxiang Wang, Yanjun Huang
AI summary
Problem
Traditional modular systems suffer from cumulative error propagation, while existing end-to-end models rely on static one-shot inference that fails to capture dynamic environmental changes and diverse driving behaviors.
Approach
The framework sequentially generates trajectory waypoints while dynamically routing scene-specific queries to specialized expert networks, enhanced by a lightweight batch reallocation strategy for efficient training.
Key results
- First integration of Mixture-of-Experts into end-to-end autonomous driving planning
- Autoregressive waypoint generation preserves temporal dependencies and handles ambiguous guidance
- Achieves 86.9 PDMS and 83.1 EPDMS on the NAVSIM dataset with a ResNet-34 backbone
- Lightweight batch reallocation strategy significantly accelerates expert network training
Why it matters
Provides a robust, adaptive planning framework that overcomes static model limitations, advancing reliable autonomous navigation for diverse real-world scenarios.
Abstract
This paper presents ARTEMIS, an end-to-end au- tonomous driving framework that combines autoregressive tra- jectory planning with Mixture-of-Experts (MoE). Traditional modular methods suffer from error propagation, while existing end-to-end models typically employ static one-shot inference paradigms that inadequately capture the dynamic changes of the environment. ARTEMIS takes a different method by generating trajectory waypoints sequentially, preserves critical temporal dependencies while dynamically routing scene-specific queries to specialized expert networks. It effectively relieves trajectory quality degradation issues encountered when guidance informa- tion is ambiguous, and overcomes the inherent representational limitations of singular network architectures when processing diverse driving scenarios. Additionally, we use a lightweight batch reallocation strategy that significantly improves the training speed of the Mixture-of-Experts model. Through experiments on the NAVSIM dataset, ARTEMIS exhibits superior competitive performance, achieving 86.9 PDMS and 83.1 EPDMS with ResNet-34 backbone, demonstrates state-of-the-art performance on multiple metrics. Code will be available under https://github. com/Lg0914/ARTEMIS.