HelixSystem2Latent<T>
Models & Types · AiDotNet.VisionLanguage.Robotics
Conditioning signal that Helix’s slow System-2 VLM produces for the fast System-1 controller.
How It Works
Per Helix (Figure AI, 2025, “Helix: A Vision-Language-Action Model for Generalist Humanoid Control”, arXiv:2502.07092) the dual-system architecture splits responsibilities:
This class captures the per-tick S2 latent (timestamp, latent tensor, freshness counter) so the dual-system runner can decide whether to re-invoke S2 or simply reuse the cached latent for the next S1 step.
Constructors
| Constructor | Summary |
|---|---|
HelixSystem2Latent(Tensor<T> latent, int producedAtTick, int validForTicks) |
Properties
| Property | Summary |
|---|---|
Tensor<T> Latent { get; } |
The latent feature vector produced by S2 — typically DecoderDim-sized. |
int ProducedAtTick { get; } |
S1 tick on which this latent was produced. |
int ValidForTicks { get; } |
Maximum number of S1 ticks this latent remains valid; after that the runner must produce a fresh one. |
Methods
| Method | Summary |
|---|---|
T L2Norm() |
Returns the L2-norm of the latent — useful for monitoring “S2 confidence” between invocations. |
bool IsStaleAt(int currentTick) |
Returns true when the latent has expired and S2 must be re-invoked. |