Learning to Navigate in Mazes with Novel Layouts using Abstract Top-down Maps
This work addresses the problem of enabling agents to navigate in unseen environments without retraining, which is incremental as it builds on existing model-based RL methods for multi-task learning.
The paper tackles zero-shot navigation in novel maze layouts using abstract top-down maps, proposing a model-based reinforcement learning approach that jointly learns a hypermodel to predict transition network weights, resulting in better adaptation and robustness to noise in the DeepMind Lab environment.
Learning navigation capabilities in different environments has long been one of the major challenges in decision-making. In this work, we focus on zero-shot navigation ability using given abstract $2$-D top-down maps. Like human navigation by reading a paper map, the agent reads the map as an image when navigating in a novel layout, after learning to navigate on a set of training maps. We propose a model-based reinforcement learning approach for this multi-task learning problem, where it jointly learns a hypermodel that takes top-down maps as input and predicts the weights of the transition network. We use the DeepMind Lab environment and customize layouts using generated maps. Our method can adapt better to novel environments in zero-shot and is more robust to noise.