Applying interpolation-constrained autoencoders to world models approach reinforcement learning

World Models Approach Reinforcement Learning helps to tackle complex problems by breaking down the learning task to Vision Model, Memory Model, and Controller Model. Variational Autoencoder (VAE) is commonly used for Vision Model. However, there has been development of other variants of Autoencoders...

Full description

Saved in:
Bibliographic Details
Main Author: Kevin Winata
Other Authors: Zinovi Rabinovich
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2021
Subjects:
Online Access:https://hdl.handle.net/10356/148004
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:World Models Approach Reinforcement Learning helps to tackle complex problems by breaking down the learning task to Vision Model, Memory Model, and Controller Model. Variational Autoencoder (VAE) is commonly used for Vision Model. However, there has been development of other variants of Autoencoders, with one prominent example is ACAI (Adversarially Constrained Autoencoder Interpolations). In this paper, we propose to substitute ACAI to VAE which might improve the performance on Open AI Car Racing environment. Unfortunately, the ingenuity of ACAI does not apply well to the Car Racing environment because of how ACAI is modeled.