planner.core.clipping_method#

Module Contents#

planner.core.clipping_method.clip_actions(actions: torch.FloatTensor, params: jacta.planner.core.parameter_container.ParameterContainer) → torch.FloatTensor#
Parameters:

  • actions (torch.FloatTensor) –

  • params (jacta.planner.core.parameter_container.ParameterContainer) –

Return type:

torch.FloatTensor

planner.core.clipping_method.box_scaling(v: torch.FloatTensor, v_min: torch.FloatTensor, v_max: torch.FloatTensor, v_mid: torch.FloatTensor | None = None) → torch.FloatTensor#

Scales vector v down to ensure that the scaled version of v (v_bar) belongs to the box [v_min, v_max]. The scaling is performed about a centerpoint v_mid. v = v_mid + n v_bar = v_mid + alpha n with alpha in [0, 1]

Parameters:

  • v (torch.FloatTensor) –

  • v_min (torch.FloatTensor) –

  • v_max (torch.FloatTensor) –

  • v_mid (Optional[torch.FloatTensor]) –

Return type:

torch.FloatTensor