Learning to Predict Without Looking Ahead: World Models Without Forward Prediction

Abstract

Much of model-based reinforcement learning involves learning a model of an agent's world, and training an agent to leverage this model to perform a task more efficiently. While these models are demonstrably useful for agents, every naturally occurring model of the world of which we are aware-e.g., a brain-arose as the byproduct of competing evolutionary pressures for survival, not minimization of a supervised forward-predictive loss via gradient descent. That useful models can arise out of the messy and slow optimization process of evolution suggests that forward-predictive modeling can arise as a side-effect of optimization under the right circumstances. Crucially, this optimization process need not explicitly be a forward-predictive loss. In this work, we introduce a modification to traditional reinforcement learning which we call observational dropout, whereby we limit the agents ability to observe the real environment at each timestep. In doing so, we can coerce an agent into learning a world model to fill in the observation gaps during reinforcement learning. We show that the emerged world model, while not explicitly trained to predict the future, can help the agent learn key skills required to perform well in its environment. Videos of our results available at https://learningtopredict.github.io/

Predictive Control Using Learned State Space Models via Rolling Horizon Evolution

“…One of the most challenging aspects of MBRL is that the model that is acquired must be sophisticated enough to capture the dynamics of the environment while being flexible and robust to potential errors. Some research has focused on producing faithful reconstructions (Kaiser et al 2020; while others have opted instead on learning to predict specific aspects of the RL framework (Schrittwieser et al 2020) or the environment (Freeman, Metz, and Ha 2019) that are thought to be necessary to perform a particular task.…”