Converting to FD plan fails when using the a-star search planner

[yijiangh]

I'm trying to use a non-default pyplanner setting, starting from the following setup:

planner = {
                'search': 'a_star', # eager | lazy | hill_climbing | a_star | random_walk | mcts
                'evaluator': 'greedy', # 'bfs' | 'uniform' | 'astar' | 'wastar2' | 'wastar3' | 'greedy'
                'heuristic': 'ff', # goal | add | ff | max | blind
                'successors': 'all', # all | random | first_goals | first_operators,
            }

However, after a plan is found, PDDLStream has a step that converts the plan back to FD format that fails due to a missing state key in the fd_action_from_py_action[action]:

pddlstream\pddlstream\language\attachments.py", line 198, in solve_pyplanners
    fd_action_from_py_action[action][state]
KeyError: State{Atom atrack(v10), Atom atrack(v11), Atom atrack(v12), Atom atrack(v13), Atom atrack(v14), Atom atrack(v15), Atom atrack(v16), Atom robottoolchangerempty(), Atom toolnotoccupiedonjoint(v10), Atom toolnotoccupiedonjoint(v11), Atom toolnotoccupiedonjoint(v12), Atom toolnotoccupiedonjoint(v13)}

I printed and found that the fd_action_from_py_action[action][state] is an empty dict! Weirdly, when using the eager search setup, this will not happen. Still looking into the cause of this problem and will report back soon.