beyretb/AnimalAI-Olympics
📦 Open Source Projectbeyretb
A benchmark competition for testing AI agents on cognitive tasks inspired by animal intelligence.
The AnimalAI-Olympics repository provides a comprehensive suite of tools for training and evaluating artificial intelligence agents in a 3D environment. Built upon the Unity game engine, the platform presents agents with a series of 'Olympics' tasks that mirror cognitive tests used in animal psychology. These tasks are designed to assess core cognitive functions such as navigation, planning, and causal reasoning in a dynamic, physics-enabled world.
Key features include a modular task design, support for deep reinforcement learning (DRL) integration, and a standardized evaluation protocol that allows researchers to compare agent performance across diverse cognitive domains. The environment is highly configurable, enabling the creation of custom scenarios that challenge agents to adapt to new situations rather than memorizing specific paths. By focusing on animal cognition, the project encourages the development of more robust, flexible, and intelligent agents capable of navigating real-world complexities.
💡Highlights
- ├─Unity-based 3D cognitive environment
- ├─Tests spatial and causal reasoning
- └─Standardized RL benchmark platform
🎯For
- ├─Reinforcement Learning Researchers
- └─Cognitive Scientists