Deep Active Inference for Pixel-Based Discrete Control: Evaluation on the Car Racing Problem

TL;DR

This paper evaluates a deep active inference agent on a pixel-based car racing task, demonstrating its ability to learn state representations and control through unsupervised learning and free energy optimization, with performance comparable to deep Q-learning.

Contribution

It introduces a deep active inference approach for pixel-based control in a complex environment and analyzes its performance and limitations.

Findings

Achieves performance comparable to deep Q-learning.

Vanilla dAIF does not outperform current world model approaches.

Identifies limitations and suggests potential architectural improvements.

Abstract

Despite the potential of active inference for visual-based control, learning the model and the preferences (priors) while interacting with the environment is challenging. Here, we study the performance of a deep active inference (dAIF) agent on OpenAI's car racing benchmark, where there is no access to the car's state. The agent learns to encode the world's state from high-dimensional input through unsupervised representation learning. State inference and control are learned end-to-end by optimizing the expected free energy. Results show that our model achieves comparable performance to deep Q-learning. However, vanilla dAIF does not reach state-of-the-art performance compared to other world model approaches. Hence, we discuss the current model implementation's limitations and potential architectures to overcome them.