Visual Analytics Dataset: Deep Reinforcement Learning for NPC Behavior in Unity ML-Agents

This dataset contains a comprehensive collection of visual analytics and evaluation metrics generated during the research on adaptive non-player character (NPC) behavior using Deep Reinforcement Learning. It illustrates the translation of the IPIP-50 psychological model into dynamic physical and semantic strategies using Curriculum Learning across various RL architectures (PPO, SAC, RND, ICM). The dataset is organized into the following analytical clusters: academic_spatial_evolution: Agent navigation trajectories and spatial occupancy heatmaps tracking movement evolution across curriculum lessons. action_correlation: Pearson correlation matrices demonstrating the direct mathematical link between IPIP-50 personality traits and specific kinematic actions. behavior_evolution: t-SNE clustering, centroid trajectories, and radar charts detailing the semantic evolution of psychological behaviors and interactions. kinematic_landscape: Dimensionality reduction (t-SNE) and zonal contours mapping the agents' physical action space and movement strategies. performance_profiling: Detailed agent performance metrics, including collision rates, time complexity, and stacked penalty decomposition (movement, rotation, neck strain, etc.). phase_space_attractors: Reward-penalty phase space trajectories illustrating learning convergence and behavioral attractors for different configurations. sensory_sequence: Temporal chronograms of sensory activation (sight, hearing, touch) and correlation between cognitive processing capacity and environmental complexity. training_dynamics: Smoothed TensorBoard evaluation metrics and mathematical learning forecasts (Linear, Logarithmic, Quadratic, and Sigmoid models).