Lecture

Spatial Memory in RL Agents

Description

This lecture explores the use of spatial and pattern memory in reinforcement learning agents through maze navigation tasks. The approach involves designing mazes of increasing complexity to test spatial orientation and path choosing abilities. The RL agent is based on a Dueling Network architecture with Double Q-learning for state updates. Results show that increasing visual landmarks improve agent performance in spatial orientation, while path choosing remains inconsistent. Limitations include the lack of environment map building and the need for better reward functions. Possible extensions involve refining starting points and orientations.

About this result
This page is automatically generated and may contain information that is not correct, complete, up-to-date, or relevant to your search query. The same applies to every other page on this website. Please make sure to verify the information with EPFL's official sources.