GNG SLAM

Goldhoorn (2006) Simulation

Nodes: 0
Edges: 0
Error: 0.00
📖 About GNG-SLAM

How it Works

This simulation demonstrates the mapping component of the GNG-SLAM algorithm (Goldhoorn et al.).

1. The Robot (Blue)

The agent moves through the environment using simulated Lidar sensors. The yellow rays measure distance to the nearest wall.

2. Sensor Hits (Red Dots)

When a ray hits a wall, it creates a "Hit Point". In standard mapping, these are just pixels. In this algorithm, these points act as training signals for the neural network.

3. The Network (Green)

A Growing Neural Gas (GNG) is an unsupervised learning network that dynamically changes its structure:

  • Adaptation: The nearest node ("winner") moves towards the sensor hit.
  • Growth: If a node accumulates too much error (distance from hits), the network spawns a new node to cover that area better.
  • Topology: Edges are created between nodes that fire together, effectively learning the "shape" of the obstacles.