CoBeL-RL

CoBeL-RL provides a closed-loop reinforcement learning simulation framework to model animal behavior and neuronal representations for computational neuroscience research.

Key Features:

• Closed-Loop Simulation Framework: Implements closed-loop interactions between an agent and environment to model dynamic behavior and learning processes.
• Virtual Environments: Supports configurable environments including T-maze and Morris water maze paradigms represented as gridworlds or detailed 3D environments with complex visual stimuli.
• Range of RL Algorithms: Includes reinforcement learning algorithms such as Dyna-Q and deep Q-networks and permits extension or integration of additional RL methods.
• Behavioral and Neural Monitoring Tools: Records behavioral outputs and unit activity within simulations for analysis of learning and neural representations.
• Fine-Grained Control: Provides programmatic hooks at key points in the closed-loop system to precisely manipulate and observe experimental variables.

Scientific Applications:

• Modeling animal behavior and learning: Simulates learning processes and behavioral responses in canonical neuroscience paradigms.
• Studying neuronal representations: Investigates how unit activity and representations emerge during reinforcement learning in simulated neural models.
• Comparative evaluation of RL models: Enables integration and comparison of RL-based models to study algorithmic and representational differences relevant to neuroscience.

Methodology:

Implements closed-loop simulations in configurable virtual environments (gridworlds to 3D visual scenes), applies RL algorithms such as Dyna-Q and deep Q-networks, and records behavioral outputs and unit activity with programmatic hooks at key closed-loop points.