https://github.com/francesco-mannella/td-0-actor-critic-reinforcement-learning-for-cue-based-localization

https://github.com/francesco-mannella/td-0-actor-critic-reinforcement-learning-for-cue-based-localization

Last synced: 21 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/francesco-mannella/td-0-actor-critic-reinforcement-learning-for-cue-based-localization
• Owner: francesco-mannella
• License: mit
• Created: 2023-11-18T10:13:30.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-12-28T11:52:16.000Z (about 1 year ago)
• Last Synced: 2024-11-07T09:23:52.719Z (2 months ago)
• Language: Python
• Size: 1.13 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
#  TD(0) Actor-Critic reinforcement learning for cue-based localization

The code is a neural network implementation of the TD(0) Actor-Critic reinforcement learning architecture. The model is based on the work in [Francesco Mannella & Gianluca Baldassarre (2006)](https://royalsocietypublishing.org/doi/10.1098/rstb.2006.1966), which aims to reproduce and understand the behaviors of domestic chicks in experiments where they are trained to locate food in different-sized arenas. The model provides insights into the chicks' navigation behaviors and suggests possible cognitive mechanisms underlying their actions. 

In the simulator, the agent sees a clear 3D view of the arena walls' edges. Movements are defined across two continuous dimensions: speed and rotation. The neural architecture is a perceptron equipped with two sets of action units — one dedicated to speed and the other to rotation — alongside an evaluation unit. The learned weights help us understand how the agent manages its actions in different visual situations.