gpc_policy/lib.rs
1//! Diffusion-based action policy implementation.
2//!
3//! Implements the generative action policy that produces candidate
4//! action sequences given observational input, using DDPM (Denoising
5//! Diffusion Probabilistic Models).
6//!
7//! # Architecture
8//!
9//! The policy uses a conditional denoising network that takes:
10//! - Noisy action sequence `[batch, pred_horizon * action_dim]`
11//! - Observation conditioning `[batch, obs_horizon * obs_dim]`
12//! - Diffusion timestep embedding
13//!
14//! And predicts the noise to remove, iteratively denoising pure
15//! Gaussian noise into a valid action sequence.
16
17pub mod network;
18pub mod policy;
19
20pub use network::{DenoisingNetwork, DenoisingNetworkConfig};
21pub use policy::{DiffusionPolicy, DiffusionPolicyConfig};