A Hierarchical Game and Deep Reinforcement Learning Approach for Incentive Mechanism Design in Autonomous‐Driving Data Trading

ABSTRACT

The rapid development of autonomous‐driving technology has generated increased demand for high‐quality, heterogeneous training data with spatiotemporal correlation; however, the current data‐trading market faces several challenges owing to unclear pricing and incentive mechanisms. This paper presents a comprehensive multi‐agent‐based incentive mechanism for autonomous‐driving data trading that addresses these critical challenges. First, we proposes a theoretical framework that represents the unique characteristics of vehicular data and complex stakeholder relationships. Then, we design a two‐layer hierarchical Stackelberg game model comprising a platform–supplier Stackelberg game and a buyer–platform game. We mathematically prove the existence of a Stackelberg equilibrium where optimal pricing strategies and data‐submission volumes maximise utilities for all participants. To address real‐world data‐trading scenarios, we implement an incentive mechanism based on deep reinforcement learning within an actor‐critic framework. Comprehensive numerical simulations across various scenarios demonstrate that the full model certifies an approximate Stackelberg equilibrium and the proposed mechanism outperforms baseline strategies both in long‐term utility and adaptability to varying demand volume. These findings provide valuable insights for designing next‐generation autonomous‐driving data marketplaces and contribute to the broader understanding of multi‐agent systems in complex trading environments.