compute_goal_distance_potentials

rlox_core::training::reward_shaping

Function compute_goal_distance_potentials 

Source 
pub fn compute_goal_distance_potentials(
    observations: &[f64],
    goal: &[f64],
    obs_dim: usize,
    goal_start: usize,
    goal_dim: usize,
    scale: f64,
) -> Result<Vec<f64>, RloxError>
Expand description

Goal-distance potential: Phi(s) = -scale * ||s[goal_slice] - goal||_2

ยงArguments

  • observations - flat (N * obs_dim) array
  • goal - target goal vector, length goal_dim
  • obs_dim - dimensionality of each observation
  • goal_start - starting index within obs where goal-relevant dims begin
  • goal_dim - number of goal-relevant dimensions
  • scale - scaling factor for the potential