Sustainable Adaptation for Autonomous Driving with the Mixture of Progressive Experts Networ

2025-02-09

AI Summary

To address the limitations of autonomous driving systems in continual learning under long-tailed data distributions and dynamic, previously unseen scenarios—namely catastrophic forgetting and constrained generalization—this paper proposes a dynamic progressive optimization framework integrating reinforcement learning with supervised learning. The core innovation is the Mixture of Progressive Experts (MoPE) network, which employs a task-aware routing mechanism for dynamic expert selection and enables progressive structural refinement, thereby balancing knowledge accumulation and model evolution. Evaluated on complex urban road simulations, the method achieves a 7.3% performance gain over behavioral cloning and demonstrates significantly enhanced robustness to novel scenarios. It effectively overcomes key generalization and stability bottlenecks inherent in conventional continual learning approaches for autonomous driving tasks.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Life-Long and Continual Learning
• Multiagent Systems: Multiagent Learning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• 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

Abstract

Learning-based autonomous driving methods require continuous acquisition of domain knowledge to adapt to diverse driving scenarios. However, due to the inherent challenges of long-tailed data distribution, current approaches still face limitations in complex and dynamic driving environments, particularly when encountering new scenarios and data. This underscores the necessity for enhanced continual learning capabilities to improve system adaptability. To address these challenges, the paper introduces a dynamic progressive optimization framework that facilitates adaptation to variations in dynamic environments, achieved by integrating reinforcement learning and supervised learning for data aggregation. Building on this framework, we propose the Mixture of Progressive Experts (MoPE) network. The proposed method selectively activates multiple expert models based on the distinct characteristics of each task and progressively refines the network architecture to facilitate adaptation to new tasks. Simulation results show that the MoPE model outperforms behavior cloning methods, achieving up to a 7.3% performance improvement in intricate urban road environments.

Problem

• Enhance continual learning in autonomous driving
• Address long-tailed data distribution challenges
• Improve adaptability in dynamic driving environments

Innovation

• Dynamic progressive optimization framework
• Mixture of Progressive Experts network
• Reinforcement and supervised learning integration