A Few Shot Adaptation of Visual Navigation Skills to New Observations using Meta-Learning

TL;DR

This paper presents a meta-learning approach that enables robots to quickly adapt their visual navigation policies to new sensors or targets using only a few training examples, improving generalization over traditional deep RL methods.

Contribution

The authors introduce a novel meta-learning algorithm with a specialized policy architecture that allows rapid adaptation of visual navigation skills to new scenarios with minimal data.

Findings

Adapts navigation policy with only three shots for unseen sensor configurations.

Effective hyperparameter analysis supports the robustness of the method.

Outperforms traditional methods in rapid adaptation scenarios.

Abstract

Target-driven visual navigation is a challenging problem that requires a robot to find the goal using only visual inputs. Many researchers have demonstrated promising results using deep reinforcement learning (deep RL) on various robotic platforms, but typical end-to-end learning is known for its poor extrapolation capability to new scenarios. Therefore, learning a navigation policy for a new robot with a new sensor configuration or a new target still remains a challenging problem. In this paper, we introduce a learning algorithm that enables rapid adaptation to new sensor configurations or target objects with a few shots. We design a policy architecture with latent features between perception and inference networks and quickly adapt the perception network via meta-learning while freezing the inference network. Our experiments show that our algorithm adapts the learned navigation policy with only three shots for unseen situations with different sensor configurations or different target colors. We also analyze the proposed algorithm by investigating various hyperparameters.