Adaptive World Models: Learning Behaviors by Latent Imagination Under Non-Stationarity
This work addresses the challenge of non-stationarity in reinforcement learning for embodied agents, representing an incremental advancement in adaptive world models.
The paper tackled the problem of enabling embodied intelligence to adapt to non-stationary environments by introducing a Hidden Parameter-POMDP formalism, which demonstrated robust behavior learning across various non-stationary RL benchmarks and unsupervised task abstraction.
Developing foundational world models is a key research direction for embodied intelligence, with the ability to adapt to non-stationary environments being a crucial criterion. In this work, we introduce a new formalism, Hidden Parameter-POMDP, designed for control with adaptive world models. We demonstrate that this approach enables learning robust behaviors across a variety of non-stationary RL benchmarks. Additionally, this formalism effectively learns task abstractions in an unsupervised manner, resulting in structured, task-aware latent spaces.