refine_plan.algorithms.refine

Functions which bring the stages of BT refinement together.

There are two key stages to refinement: 1. State space extraction and option learning. 2. Semi-MDP construction and solution, and policy->BT conversion.

These two stages are represented with different functions.

Author: Charlie Street Owner: Charlie Street

Module Contents

Functions

synthesise_bt_from_options(sf_list, option_list, labels)

Synthesise a BT using options learned from the initial BT.
refine_plan.algorithms.refine.synthesise_bt_from_options(sf_list, option_list, labels, initial_state=None, prism_prop='Rmax=?[F "goal"]', default_action='None', out_file=None, bt_converter='normal')

Synthesise a BT using options learned from the initial BT.

Parameters:

  • sf_list – A list of StateFactor objects for the semi-MDP

  • option_list – A list of Option objects which capture robot capabilities

  • labels – A list of Label objects within the semi-MDP

  • initial_state – The semi-MDP initial state (if there is one)

  • prism_prop – The PRISM property to solve for (passed to Storm)

  • default_action – The action which replaces any None actions in a policy

  • out_file – Optional. The path for the final BT XML file

  • bt_converter – Can be normal (PolicyBTConverter) or simple (SimplePolicyBTConverter)

Returns:

The final refined BT