A Design of a CaSSLe-Based Offline Continual Reinforcement Learning Model in Autonomous Driving Edge Devices

Key Points

• This research aims to enhance adaptation in autonomous driving systems using continual reinforcement learning algorithms.
• Introduced the CaSSLe algorithm for continual learning to a Double DQN model.
• Developed AP-CaSSLe for dynamic adjustment of data augmentation strength.
• Compared AP-CaSSLe's performance with standard Double DQN in edge device environments.
• AP-CaSSLe achieved an accuracy improvement of 58.9% over the standard Double DQN.
• Indicates a significant performance enhancement potential in autonomous driving edge applications.