Module neural 

Small CPU supervised networks with portable JSON weights and a planar steering component. This is imitation/regression training, not reinforcement learning, navigation or NVIDIA inference.

Structs

Example

One supervised training pair. Normalize features and targets before training.

Network

One tanh hidden layer and a linear output layer, stored as row-major JSON weights.

NeuralAgent

Small supervised steering controller. Inputs are normalized world X/Z offsets to a named target. Outputs are planar movement X/Z. Does not perform pathfinding or control dynamic rigid bodies.

Functions

steering_examples

Reproducible steering examples: normalized X/Z offsets mapped to desired X/Z movement.

update

Advances enabled, valid steering agents. Invalid models/settings leave entities untouched.