Bidirectional Fast Downward: Satisficing Bidirectional Heuristic Search Planner

Ryo Kuroiwa and Alex Fukunaga. Front-to-Front Heuristic Search for Satisficing Classical Planning. IJCAI 2020.

Build

$ docker build . -t bfd
$ docker run 

Planners

Attach to the container and execute following commands. Replace domain.pddl and problem.pddl with domain and problem files that you want to solve, or specify a single SAS+ file.

TTBS

$ python3 fast-downward.py domain.pddl problem.pddl --search 'bidirectional_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=goal)), front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=initial)), reeval=NOT_SIMILAR, d_node_type=TTBS)'

DNR(PP84) with n=75

$ python3 fast-downward.py domain.pddl problem.pddl --search bidirectional_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=goal)), front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=initial)), reeval=NOT_SIMILAR, d_node_type=MAX_G, max_steps=75) ```

DNR(OT)

$ python3 fast-downward.py domain.pddl problem.pddl --search bidirectional_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=goal)), front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=initial)), reeval=ALL, d_node_type=TTBS) ```

DNR(re)

$ python3 fast-downward.py domain.pddl problem.pddl --search bidirectional_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=goal)), front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, fall_back_to=initial)), reeval=NOT_SIMILAR, d_node_type=MAX_G) ```

SFBS

$ python3 fast-downward.py domain.pddl problem.pddl --search 'eager_sfbs(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true)))'

biFD

$ python3 fast-downward.py domain.pddl problem.pddl --search 'bidirectional_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_goal=true)), front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_initial=true)))

BDD

$ python3 fast-downward.py domain.pddl problem.pddl --search 'bidirectional_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_goal=true)), front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_initial=true)), bdd=true)

BGG

$ python3 fast-downward.py domain.pddl problem.pddl --search 'bidirectional_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_goal=true)), front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_initial=true))), bgg_eval=front_to_front_hmax(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_initial=true), use_bgg=true)

FDr

$  python3 fast-downward.py domain.pddl problem.pddl --search 'regression_eager_greedy(front_to_front_single(front_to_front_ff(transform=adapt_costs(cost_type=ONE, transform=partial_state()), partial_state=true, cache_initial=true)))'

FD

$  python3 fast-downward.py domain.pddl problem.pddl --search'eager_greedy([ff(transform=adapt_costs(ONE))])