An interactive 3D visualization of a motion autoencoder, showing animated humanoid skeletons that transition between poses in real time. The scene renders skeletal figures with colored joint connections against a dark fogged background, with camera controls allowing orbit around the models. A web-based interface loads three ONNX models (motion VAE, target pose, and joystick) and executes them via ONNX.js with a WebGL backend, while Three.js handles the 3D scene rendering and animation loop. A dat.gui control panel provides parameter adjustment for exploring the latent motion space.