Separated Models for Improved Generalization in Visual-Based Reinforcement Learning
The authors propose a novel method called SMG (Separated Models for Generalization) that leverages image reconstruction with separated model branches and consistency losses to improve the generalization ability of reinforcement learning agents in unseen environments, particularly those with visual background variations.
Training on More Reachable Tasks Improves Generalization in Reinforcement Learning, Even to Unreachable Tasks
Training reinforcement learning agents on a wider variety of reachable tasks, achieved by incorporating an initial exploration phase at the beginning of each training episode, leads to improved generalization, even to unreachable tasks that share no states or rewards with the training set.
Improving Generalization in Reinforcement Learning through Generative Learning
The author explores the use of imagination-based reinforcement learning to improve generalization in RL agents by generating dream-like trajectories. By leveraging generative augmentations, the method shows superior performance in sparsely rewarded environments.
© 2024 by Linnk AI