# Learning Manipulation Skills Via Hierarchical Spatial Attention

## TL;DR

This paper introduces a hierarchical spatial attention mechanism for robotic manipulation, enabling robots to focus on relevant areas and learn generalizable skills despite partial observability, validated through real-robot experiments.

## Contribution

It proposes a hierarchical attention approach that simplifies learning attention policies and demonstrates effective manipulation skills in real-world tasks.

## Key findings

- Hierarchical spatial attention improves focus and learning efficiency.
- Q-learning can find optimal policies despite partial observability.
- Real-robot experiments confirm practical applicability.

## Abstract

Learning generalizable skills in robotic manipulation has long been challenging due to real-world sized observation and action spaces. One method for addressing this problem is attention focus -- the robot learns where to attend its sensors and irrelevant details are ignored. However, these methods have largely not caught on due to the difficulty of learning a good attention policy and the added partial observability induced by a narrowed window of focus. This article addresses the first issue by constraining gazes to a spatial hierarchy. For the second issue, we identify a case where the partial observability induced by attention does not prevent Q-learning from finding an optimal policy. We conclude with real-robot experiments on challenging pick-place tasks demonstrating the applicability of the approach.