Reinformed Dreamer: An Asymmetric World Model Efficiently Trained through Latent Guidance
2026-07-28 • Machine Learning
Machine Learning
AI summaryⓘ
The authors explore how reinforcement learning (RL) can improve by using extra information during training, not just rewards. They focus on a way of learning called asymmetric reinforcement learning, which helps models learn better by having access to privileged or extra state information. Specifically, they study an existing method called Informed Dreamer and find it has some limits in how it learns from this extra information. To fix this, they create a new method called Reinformed Dreamer that uses a new way of learning representations called latent guidance. Their tests show that Reinformed Dreamer reliably improves performance compared to previous methods.
reinforcement learningasymmetric reinforcement learningmodel-based learningprivileged informationlatent representationInformed DreamerReinformed Dreamerpartial observabilitylatent guidance
Authors
Gaspard Lambrechts, Adrien Bolland, Daniel Ebi, Damien Ernst
Abstract
Much like humans benefit from guidance while learning, reinforcement learning algorithms may benefit from additional supervision beyond rewards. Leveraging additional information during training to learn better representations and behaviors has been the focus of asymmetric reinforcement learning. This learning paradigm has proven effective under partial observability when additional state information is available, but also under full observability when more refined state information is available. Focusing on model-based reinforcement learning, we study the effect of asymmetric learning on observation representations and on privileged information representations. First, we identify a limitation in the privileged information representations learned by an asymmetric model-based algorithm known as the Informed Dreamer. Then, we propose a novel asymmetric representation learning objective using latent guidance, resulting in a new algorithm called the Reinformed Dreamer. Experiments across several benchmarks show a more consistent improvement over Dreamer than previous asymmetric approaches.