Struct HERBuffer 

pub struct HERBuffer { /* private fields */ }

Hindsight Experience Replay buffer.

Stores transitions with goal information and performs goal relabeling during sampling. The obs vector layout is: [obs_core | achieved_goal | desired_goal | ...]

Implementations

impl HERBuffer

pub fn new( capacity: usize, obs_dim: usize, act_dim: usize, goal_dim: usize, achieved_goal_start: usize, desired_goal_start: usize, strategy: HERStrategy, goal_tolerance: f32, ) -> Self

Create a new HER buffer.

Arguments

• capacity - maximum transitions
• obs_dim - full observation dimension (includes goal components)
• act_dim - action dimension
• goal_dim - goal vector dimension
• achieved_goal_start - index within obs where achieved goal starts
• desired_goal_start - index within obs where desired goal starts
• strategy - relabeling strategy
• goal_tolerance - tolerance for sparse reward computation

pub fn push_slices( &mut self, obs: &[f32], next_obs: &[f32], action: &[f32], reward: f32, terminated: bool, truncated: bool, ) -> Result<(), RloxError>

Push a single transition, notifying the episode tracker.

pub fn sample_with_relabeling( &self, batch_size: usize, her_ratio: f32, seed: u64, ) -> Result<SampledBatch, RloxError>

Sample a batch with HER relabeling.

her_ratio controls the fraction of samples that get relabeled goals. The remaining samples use their original goals.

pub fn compute_relabel_indices( &self, episode: &EpisodeMeta, transition_offset: usize, seed: u64, ) -> Vec<usize>

Compute relabeling indices for a given episode and transition.

Returns indices (offsets within the episode) to use as substitute goals.

pub fn len(&self) -> usize

Number of valid transitions currently stored.

pub fn is_empty(&self) -> bool

Whether the buffer is empty.

pub fn num_complete_episodes(&self) -> usize

Number of complete episodes currently tracked.

Trait Implementations

impl Debug for HERBuffer

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.