Skip to content

DiscreteTimeStateEvolution

Discrete-time state evolution via Markov transition distributions.

The next state is drawn from a conditional distribution given the current state, control, and time indices:

\[ x_{t_{k+1}} \sim p\left(x_{t_{k+1}} \mid x_{t_k}, u_{t_k}, t_k, t_{k+1}\right) \]

Implementations must return a NumPyro-compatible distribution (e.g., numpyro.distributions.Distribution) that can be sampled and evaluated.

Parameters:

Name Type Description Default
x State

Current state \(x \in \mathbb{R}^{d_x}\).

required
u Control | None

Current control input or None.

required
t_now Time

Current time index \(t_k\).

required
t_next Time

Next time index \(t_{k+1}\) (for non-uniform sampling or continuous-time embeddings).

required

Returns:

Name Type Description
DistributionT

Distribution over the next state \(x_{t_{k+1}}\). In practice this should be a numpyro.distributions.Distribution instance.