IIntrinsicRewardModule<T>
Interfaces · AiDotNet.ReinforcementLearning.IntrinsicMotivation
Produces an intrinsic (curiosity) reward for a state, driving a reinforcement-learning agent toward novel states even when the environment’s extrinsic reward is sparse.
Known implementations: RandomNetworkDistillation<T>
For Beginners
Some environments only reward the agent rarely (e.g. at the very end of a long task). Without a signal in between, the agent flails. An intrinsic-reward module gives the agent a small bonus for visiting unfamiliar states, so it keeps exploring purposefully until it finds the real reward. The bonus shrinks as a state becomes familiar.
Methods
| Method | Summary |
|---|---|
T ComputeIntrinsicReward(Vector<T> state) |
Returns the intrinsic novelty reward for state (larger = more novel). |
void Reset() |
Resets any per-episode state. |
void Update(Vector<T> state) |
Learns from a visited state so it becomes less novel next time. |