
Combining disentangled representation learning with associative memory enables vision-based reinforcement learning agents to achieve zero-shot generalization on unseen task variations without relying on data augmentation.


coremsg

associative-latent-disentanglement-alda-for-zero-shot-generalization-in-vision-based-reinforcement-learning


Associative Latent Disentanglement (ALDA) for Zero-Shot Generalization in Vision-Based Reinforcement Learning


title_rewrite


Jointly learning context representations and policy enables improved zero-shot generalization in reinforcement learning.


improving-zero-shot-generalization-in-reinforcement-learning-by-learning-behavior-specific-context-representations


Improving Zero-Shot Generalization in Reinforcement Learning by Learning Behavior-Specific Context Representations
