Goal-Aware Prediction: Learning to Model What Matters

TL;DR

This paper introduces a goal-aware prediction method that directs learned dynamics models to focus on task-relevant information, improving performance in vision-based control tasks without requiring reward labels.

Contribution

It proposes a self-supervised approach to make dynamics models aware of task relevance, aligning their predictions more closely with downstream planning and policy objectives.

Findings

Outperforms standard task-agnostic models in vision-based control tasks

Effectively models relevant scene parts conditioned on goals

Does not require reward functions or image labels

Abstract

Learned dynamics models combined with both planning and policy learning algorithms have shown promise in enabling artificial agents to learn to perform many diverse tasks with limited supervision. However, one of the fundamental challenges in using a learned forward dynamics model is the mismatch between the objective of the learned model (future state reconstruction), and that of the downstream planner or policy (completing a specified task). This issue is exacerbated by vision-based control tasks in diverse real-world environments, where the complexity of the real world dwarfs model capacity. In this paper, we propose to direct prediction towards task relevant information, enabling the model to be aware of the current task and encouraging it to only model relevant quantities of the state space, resulting in a learning objective that more closely matches the downstream task. Further, we do so in an entirely self-supervised manner, without the need for a reward function or image labels. We find that our method more effectively models the relevant parts of the scene conditioned on the goal, and as a result outperforms standard task-agnostic dynamics models and model-free reinforcement learning.