Skip to content
Snippets Groups Projects
README.md 939 B
Newer Older
# Leaving The Navmesh: An Ablative Analysis of Deep Reinforcement Learning for Complex Navigation in 3D Virtual Environments
Electronic Supplementary Information for the submission "Leaving the NavMesh: An Ablative Analysis of Deep Reinforcement Learning for Complex Navigation in 3D Virtual Environments" in the Australasian Joint Conference on Artificial Intelligence 2023.

This ESM contains the following information:
1. Hyper-parameter details for the implemented SAC algorithm, the CNN architecture, and the environment.
2. Implementation details for the Poisson Disk Sampling Method utilized to generate start/goal locations for the studied environments.
3. A detailed description of the implemented Actor-Critic architecture.
4. Action Repetition Ablation Study
5. Euclidean Reward Scaling Factor Ablation Study
6. Occupancy Grid Map Ablation study for 1) The OGM sensor sample size; and 2) The Cell Size of the voxels in the OGM.