def rmpyl_icaps14():
"""
Example from (Santana & Williams, ICAPS14).
"""
prog = RMPyL()
prog *= prog.decide(
{'name':'transport-choice','domain':['Bike','Car','Stay'],
'utility':[100,70,0]},
prog.observe(
{'name':'slip','domain':[True,False],
'ctype':'probabilistic','probability':[0.051,1.0-0.051]},
prog.sequence(Episode(action='(ride-bike)',
duration={'ctype':'controllable','lb':15,'ub':25}),
Episode(action='(change)',
duration={'ctype':'controllable','lb':20,'ub':30})),
Episode(action='(ride-bike)',duration={'ctype':'controllable','lb':15,'ub':25})),
prog.observe(
{'name':'accident','domain':[True,False],
'ctype':'probabilistic','probability':[0.013,1.0-0.013]},
prog.sequence(Episode(action='(tow-vehicle)',
duration={'ctype':'controllable','lb':30,'ub':90}),
Episode(action='(cab-ride)',
duration={'ctype':'controllable','lb':10,'ub':20})),
Episode(action='(drive)',duration={'ctype':'controllable','lb':10,'ub':20})),
Episode(action='(stay)'))
prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=30.0)
return prog
def rmpyl_nested_uav():
hello = UAV('hello')
uav = UAV('uav')
prog = RMPyL()
prog.plan = prog.sequence(
hello.scan(), uav.scan(),
prog.decide(
{
'name': 'UAV-choice',
'domain': ['Hello', 'UAV'],
'utility': [7, 5]
},
prog.sequence(
hello.fly(),
prog.observe(
{
'name': 'hello-success',
'domain': ['Success', 'Failure'],
'ctype': 'probabilistic',
'probability': [0.8, 0.2]
},
prog.decide(
{
'name': 'hello-assert-success',
'domain': ['Success'],
'utility': [10]
}, hello.stop()),
prog.decide(
{
'name': 'hello-assert-failure',
'domain': ['Failure'],
'utility': [0]
}, hello.stop()))),
prog.sequence(
uav.fly(),
prog.observe(
{
'name': 'uav-success',
'domain': ['Success', 'Failure'],
'ctype': 'probabilistic',
'probability': [0.95, 0.05]
},
prog.decide(
{
'name': 'uav-assert-success',
'domain': ['Success'],
'utility': [10]
}, uav.stop()),
prog.decide(
{
'name': 'uav-assert-failure',
'domain': ['Failure'],
'utility': [0]
}, uav.stop())))))
return prog
def rmpyl_original_verbose(hello,uav):
"""
Implementation of the original RMPL using a more verbose syntax and adding
a chance constraint.
##### Original RMPL
class UAV {
value on;
value off;
primitive method fly() [3,10];
primitive method scan() [1,10];
}
class Main {
UAV helo;
UAV uav;
method run () {
[0, 18] sequence {
parallel {
sequence {
helo.scan();
helo.fly();
}
sequence {
uav.fly();
uav.scan();
}
}
choose {
with reward: 5 {helo.fly();}
with reward: 7 {uav.fly();}
}
}
}
}
"""
prog = RMPyL()
prog.plan = prog.sequence(
prog.parallel(
prog.sequence(
hello.scan(),
hello.fly()),
prog.sequence(
uav.fly(),
uav.scan())),
prog.decide({'name':'UAV-choice','domain':['Hello','UAV'],'utility':[5,7]},
hello.fly(),
uav.fly()))
overall_tc = prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=18.0)
cc_time = ChanceConstraint(constraint_scope=[overall_tc],risk=0.1)
prog.add_chance_constraint(cc_time)
return prog
def nominal_case(blocks, time_window=-1, dur_dict=None):
"""
Nominal case, where the robot observes what the human has already completed,
and acts accordingly
"""
agent = 'Baxter'
manip = 'BaxterRight'
prog = RMPyL(name='run()')
prog *= prog.sequence(
say('Should I start?'),
prog.observe(
{
'name': 'ask-human',
'ctype': 'probabilistic',
'domain': ['YES', 'NO'],
'probability': [0.9, 0.1]
}, observe_decide_act(prog, blocks, manip, agent, dur_dict),
say('All done!')))
if time_window > 0.0:
prog.add_overall_temporal_constraint(ctype='controllable',
lb=0.0,
ub=time_window)
return prog
def rmpyl_parallel_uav():
hello = UAV('hello')
uav = UAV('uav')
prog = RMPyL()
prog.plan = prog.parallel(
prog.sequence(
prog.decide(
{
'name': 'hello-action',
'domain': ['Fly', 'Scan'],
'utility': [0, 1]
}, hello.fly(), hello.scan()),
prog.decide(
{
'name': 'hello-action',
'domain': ['Fly', 'Scan'],
'utility': [0, 1]
}, hello.fly(), hello.scan()),
prog.decide(
{
'name': 'hello-action',
'domain': ['Fly', 'Scan'],
'utility': [0, 1]
}, hello.fly(), hello.scan())),
prog.sequence(
prog.decide(
{
'name': 'uav-action',
'domain': ['Fly', 'Scan'],
'utility': [0, 1]
}, uav.fly(), uav.scan()),
prog.decide(
{
'name': 'uav-action',
'domain': ['Fly', 'Scan'],
'utility': [0, 1]
}, uav.fly(), uav.scan()),
prog.decide(
{
'name': 'uav-action',
'domain': ['Fly', 'Scan'],
'utility': [0, 1]
}, uav.fly(), uav.scan())))
return prog
def rmpyl_nested_uav():
hello = UAV('hello')
uav = UAV('uav')
prog = RMPyL()
prog.plan = prog.sequence(
hello.scan(),
uav.scan(),
prog.decide(
{'name':'UAV-choice','domain':['Hello','UAV'],
'utility':[7,5]},
prog.sequence(
hello.fly(),
prog.observe(
{'name':'hello-success','domain':['Success','Failure'],
'ctype':'probabilistic','probability':[0.8,0.2]},
prog.decide(
{'name':'hello-assert-success',
'domain':['Success'],
'utility':[10]},
hello.stop()),
prog.decide(
{'name':'hello-assert-failure',
'domain':['Failure'],
'utility':[0]},
hello.stop()))),
prog.sequence(
uav.fly(),
prog.observe(
{'name':'uav-success','domain':['Success','Failure'],
'ctype':'probabilistic','probability':[0.95,0.05]},
prog.decide(
{'name':'uav-assert-success',
'domain':['Success'],
'utility':[10]},
uav.stop()),
prog.decide(
{'name':'uav-assert-failure',
'domain':['Failure'],
'utility':[0]},
uav.stop())))))
return prog
def rmpyl_episode_ids(hello,uav):
"""Example of how episode ID's can be used to retrieve them."""
prog = RMPyL()
first_uav_seq = prog.sequence(uav.scan(),uav.fly(),id='uav-1-seq')
second_uav_seq = prog.sequence(uav.scan(),uav.fly(),id='uav-2-seq')
first_hello_seq = prog.sequence(hello.scan(),hello.fly(),id='hello-1-seq')
second_hello_seq = prog.sequence(hello.scan(),hello.fly(),id='hello-2-seq')
prog *= prog.parallel(prog.sequence(first_uav_seq,second_uav_seq,id='uav-seqs'),
prog.sequence(first_hello_seq,second_hello_seq,id='hello-seqs'),id='par-seqs')
#This could have been accomplished much more easily by using the sequence
#variables directly, but I wanted to show how episodes can be retrieved by
#ID.
tc1 = TemporalConstraint(start=prog.episode_by_id('uav-1-seq').end,
end=prog.episode_by_id('hello-2-seq').start,
ctype='controllable',lb=2.0,ub=3.0)
tc2 = TemporalConstraint(start=prog.episode_by_id('hello-1-seq').end,
end=prog.episode_by_id('uav-2-seq').start,
ctype='controllable',lb=0.5,ub=1.0)
prog.add_temporal_constraint(tc1)
prog.add_temporal_constraint(tc2)
return prog
def pysat_planner(dom_file,prob_file,max_steps,duration_func):
"""
Uses PySAT as the planner.
"""
py_sat = PySAT(dom_file,prob_file,precompute_steps=max_steps,remove_static=True,
write_dimacs=False,verbose=True)
domain,problem,task = model_parser(dom_file,prob_file,remove_static=True)
print('\n##### Determining optimal plan length!\n')
start = time.time()
min_steps = len(task.goals-task.initial_state)
plans = py_sat.plan(task.initial_state,task.goals,time_steps=max_steps,
find_shortest=True,min_steps=min_steps)
elapsed = time.time()-start
print('\n##### All solving took %.4f s'%(elapsed))
if len(plans)>0:
plan = plans[0]
print('\n##### Plan found!\n')
for t,action in enumerate(plan):
print('%d: %s'%(t,action))
prog = RMPyL(name='run()')
if duration_func!=None:
prog.plan = prog.sequence(*[Episode(start=Event(name='start-of-'+op),
end=Event(name='end-of-'+op),
action=op,
duration=duration_func(op)) for op in plan])
else:
prog.plan = prog.sequence(*[Episode(start=Event(name='start-of-'+op),
end=Event(name='end-of-'+op),
action=op) for op in plan])
else:
prog = None
return {'policy':None,'explicit':None,'performance':None,'rmpyl':prog}
def rmpyl_parallel_uav():
hello = UAV('hello')
uav = UAV('uav')
prog = RMPyL()
prog.plan = prog.parallel(
prog.sequence(
prog.decide({'name':'hello-action','domain':['Fly','Scan'],
'utility':[0,1]},
hello.fly(),
hello.scan()),
prog.decide({'name':'hello-action','domain':['Fly','Scan'],
'utility':[0,1]},
hello.fly(),
hello.scan()),
prog.decide({'name':'hello-action','domain':['Fly','Scan'],
'utility':[0,1]},
hello.fly(),
hello.scan())
),
prog.sequence(
prog.decide({'name':'uav-action','domain':['Fly','Scan'],
'utility':[0,1]},
uav.fly(),
uav.scan()),
prog.decide({'name':'uav-action','domain':['Fly','Scan'],
'utility':[0,1]},
uav.fly(),
uav.scan()),
prog.decide({'name':'uav-action','domain':['Fly','Scan'],
'utility':[0,1]},
uav.fly(),
uav.scan())
))
return prog
def nominal_case(blocks):
"""
Nominal case, where the robot observes what the human has already completed,
and acts accordingly
"""
agent='Baxter'
manip='BaxterRight'
prog = RMPyL(name='run()')
prog *= prog.sequence(say('Should I start?'),
prog.observe({'name':'observe-human-%d'%(len(blocks)),
'ctype':'uncontrollable',
'domain':['YES','NO']},
observe_and_act(prog,blocks,manip,agent),
say('All done!')))
return prog
def rmpyl_uav():
hello = UAV('hello')
uav = UAV('uav')
prog = RMPyL()
# prog *= hello.fly()
prog.plan = prog.sequence(
hello.scan(), hello.fly(), uav.fly(), uav.scan(),
prog.decide(
{
'name': 'UAV-choice',
'domain': ['Hello', 'UAV'],
'utility': [5, 7]
}, hello.fly(), uav.fly()))
prog.add_overall_temporal_constraint(ctype='controllable', lb=0.0, ub=18.0)
return prog
def rmpyl_uav():
hello = UAV('hello')
uav = UAV('uav')
prog = RMPyL()
# prog *= hello.fly()
prog.plan = prog.sequence(
hello.scan(),
hello.fly(),
uav.fly(),
uav.scan(),
prog.decide({'name':'UAV-choice','domain':['Hello','UAV'],
'utility':[5,7]},
hello.fly(),
uav.fly()))
prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=18.0)
return prog
def nominal_case(blocks,time_window=-1,dur_dict=None):
"""
Nominal case, where the robot observes what the human has already completed,
and acts accordingly
"""
agent='Baxter'
manip='BaxterRight'
prog = RMPyL(name='run()')
prog *= prog.sequence(say('Should I start?'),
prog.observe({'name':'ask-human',
'ctype':'probabilistic',
'domain':['YES','NO'],
'probability':[0.9,0.1]},
observe_decide_act(prog,blocks,manip,agent,dur_dict),
say('All done!')))
if time_window>0.0:
prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=time_window)
return prog
for t, op in enumerate(plan):
print('%d: %s' % (t, op))
elapsed = time.time() - start
print('\n##### All solving took %.4f s' % (elapsed))
prog = RMPyL(name='run()')
pddl_episodes = [
Episode(id=make_episode_id(t, op),
start=Event(name='start-of-%d-%s' % (t, op)),
end=Event(name='end-of-%d-%s' % (t, op)),
action=op,
duration=rss_duration_model_func(op))
for t, op in enumerate(plan)
]
prog.plan = prog.sequence(*pddl_episodes)
# prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=2000.0)
#Adds temporal window to the plan
for t, op in enumerate(plan):
bounds, tc_type = rss_time_window_model_func(op)
for tc in time_window_constraints(
tc_type, bounds, prog.first_event,
prog.episode_by_id(make_episode_id(t, op))):
prog.add_temporal_constraint(tc)
#Dummy episodes that enable transmissions
activation_episodes = []
activation_tcs = []
global_start = Event(name='global-start')
for op_name, op_param_dict in time_windows['time_windows'].items():
start = time.time()
sat_plans = py_sat.plan(task.initial_state, task.goals, time_steps=18)
elapsed = time.time() - start
print('\n##### All solving took %.4f s' % (elapsed))
if len(sat_plans) > 0:
plan = sat_plans[0]
print('\n##### Plan found!\n')
for t, action in enumerate(plan):
print('%d: %s' % (t, action))
prog = RMPyL(name='run()')
prog.plan = prog.sequence(*[
Episode(start=Event(name='start-of-' + op),
end=Event(name='end-of-' + op),
action=op,
duration=rss_duration_func(op)) for op in plan
])
prog.add_overall_temporal_constraint(ctype='controllable',
lb=0.0,
ub=2000.0)
prog.to_ptpn(filename='rss_pysat_before_stnu_reform.tpn')
paris = PARIS()
risk_bound, sc_sched = paris.stnu_reformulation(prog,
makespan=True,
cc=0.001)
if risk_bound != None:
risk_bound = min(risk_bound, 1.0)
print(
'\nSuccessfully performed STNU reformulation with scheduling risk %f %%!'
def rmpyl_breakfast():
"""
Example from (Levine & Williams, ICAPS14).
"""
#Actions that Alice performs
get_mug_ep = Episode(action='(get alice mug)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
get_glass_ep = Episode(action='(get alice glass)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
make_cofee_ep = Episode(action='(make-coffee alice)',duration={'ctype':'controllable','lb':3.0,'ub':5.0})
pour_cofee_ep = Episode(action='(pour-coffee alice mug)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
pour_juice_glass = Episode(action='(pour-juice alice glass)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
get_bagel_ep = Episode(action='(get alice bagel)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
get_cereal_ep = Episode(action='(get alice cereal)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
toast_bagel_ep = Episode(action='(toast alice bagel)',duration={'ctype':'controllable','lb':3.0,'ub':5.0})
add_cheese_bagel_ep = Episode(action='(add-cheese alice bagel)',duration={'ctype':'controllable','lb':1.0,'ub':2.0})
mix_cereal_ep = Episode(action='(mix-cereal alice milk)',duration={'ctype':'controllable','lb':1.0,'ub':2.0})
#Actions that the robot performs
get_grounds_ep = Episode(action='(get grounds robot)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
get_juice_ep = Episode(action='(get juice robot)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
get_milk_ep = Episode(action='(get milk robot)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
get_cheese_ep = Episode(action='(get cheese robot)',duration={'ctype':'controllable','lb':0.5,'ub':1.0})
prog = RMPyL()
prog *= prog.sequence(
prog.parallel(
prog.observe(
{'name':'observe-utensil','domain':['Mug','Glass'],'ctype':'uncontrollable'},
get_mug_ep,
get_glass_ep,
id='observe-utensil-ep'),
prog.decide(
{'name':'choose-beverage-ingredient','domain':['Grounds','Juice'],'utility':[0,0]},
get_grounds_ep,
get_juice_ep,
id='choose-beverage-ingredient-ep')),
prog.observe(
{'name':'observe-alice-drink','domain':['Coffee','Juice'],'ctype':'uncontrollable'},
prog.sequence(make_cofee_ep,pour_cofee_ep),
pour_juice_glass,
id='observe-alice-drink-ep'),
prog.parallel(
prog.observe(
{'name':'observe-food','domain':['Bagel','Cereal'],'ctype':'uncontrollable'},
get_bagel_ep,
get_cereal_ep,
id='observe-food-ep'),
prog.decide(
{'name':'choose-food-ingredient','domain':['Milk','Cheese'],'utility':[0,0]},
get_milk_ep,
get_cheese_ep,
id='choose-food-ingredient-ep'),
id='parallel-food-ep'),
prog.observe(
{'name':'observe-alice-food','domain':['Bagel','Cereal'],'ctype':'uncontrollable'},
prog.sequence(toast_bagel_ep,add_cheese_bagel_ep),
mix_cereal_ep),
id='breakfast-sequence')
extra_tcs = [TemporalConstraint(start=prog.episode_by_id('breakfast-sequence').start,
end=prog.episode_by_id('observe-utensil-ep').start,
ctype='controllable',lb=0.0,ub=0.0),
TemporalConstraint(start=prog.episode_by_id('breakfast-sequence').start,
end=prog.episode_by_id('choose-beverage-ingredient-ep').start,
ctype='controllable',lb=0.2,ub=0.3),
TemporalConstraint(start=prog.episode_by_id('parallel-food-ep').start,
end=prog.episode_by_id('observe-food-ep').start,
ctype='controllable',lb=0.0,ub=0.0),
TemporalConstraint(start=prog.episode_by_id('parallel-food-ep').start,
end=prog.episode_by_id('choose-food-ingredient-ep').start,
ctype='controllable',lb=0.2,ub=0.3)]
for tc in extra_tcs:
prog.add_temporal_constraint(tc)
prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=7.0)
prog.simplify_temporal_constraints()
return prog
def rmpyl_breakfast():
"""
Example from (Levine & Williams, ICAPS14).
"""
#Actions that Alice performs
get_mug_ep = Episode(action='(get alice mug)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
get_glass_ep = Episode(action='(get alice glass)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
make_cofee_ep = Episode(action='(make-coffee alice)',
duration={
'ctype': 'controllable',
'lb': 3.0,
'ub': 5.0
})
pour_cofee_ep = Episode(action='(pour-coffee alice mug)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
pour_juice_glass = Episode(action='(pour-juice alice glass)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
get_bagel_ep = Episode(action='(get alice bagel)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
get_cereal_ep = Episode(action='(get alice cereal)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
toast_bagel_ep = Episode(action='(toast alice bagel)',
duration={
'ctype': 'controllable',
'lb': 3.0,
'ub': 5.0
})
add_cheese_bagel_ep = Episode(action='(add-cheese alice bagel)',
duration={
'ctype': 'controllable',
'lb': 1.0,
'ub': 2.0
})
mix_cereal_ep = Episode(action='(mix-cereal alice milk)',
duration={
'ctype': 'controllable',
'lb': 1.0,
'ub': 2.0
})
#Actions that the robot performs
get_grounds_ep = Episode(action='(get grounds robot)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
get_juice_ep = Episode(action='(get juice robot)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
get_milk_ep = Episode(action='(get milk robot)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
get_cheese_ep = Episode(action='(get cheese robot)',
duration={
'ctype': 'controllable',
'lb': 0.5,
'ub': 1.0
})
prog = RMPyL()
prog *= prog.sequence(
prog.parallel(
prog.observe(
{
'name': 'observe-utensil',
'domain': ['Mug', 'Glass'],
'ctype': 'uncontrollable'
},
get_mug_ep,
get_glass_ep,
id='observe-utensil-ep'),
prog.decide(
{
'name': 'choose-beverage-ingredient',
'domain': ['Grounds', 'Juice'],
'utility': [0, 0]
},
get_grounds_ep,
get_juice_ep,
id='choose-beverage-ingredient-ep')),
prog.observe(
{
'name': 'observe-alice-drink',
'domain': ['Coffee', 'Juice'],
'ctype': 'uncontrollable'
},
prog.sequence(make_cofee_ep, pour_cofee_ep),
pour_juice_glass,
id='observe-alice-drink-ep'),
prog.parallel(prog.observe(
{
'name': 'observe-food',
'domain': ['Bagel', 'Cereal'],
'ctype': 'uncontrollable'
},
get_bagel_ep,
get_cereal_ep,
id='observe-food-ep'),
prog.decide(
{
'name': 'choose-food-ingredient',
'domain': ['Milk', 'Cheese'],
'utility': [0, 0]
},
get_milk_ep,
get_cheese_ep,
id='choose-food-ingredient-ep'),
id='parallel-food-ep'),
prog.observe(
{
'name': 'observe-alice-food',
'domain': ['Bagel', 'Cereal'],
'ctype': 'uncontrollable'
}, prog.sequence(toast_bagel_ep, add_cheese_bagel_ep),
mix_cereal_ep),
id='breakfast-sequence')
extra_tcs = [
TemporalConstraint(
start=prog.episode_by_id('breakfast-sequence').start,
end=prog.episode_by_id('observe-utensil-ep').start,
ctype='controllable',
lb=0.0,
ub=0.0),
TemporalConstraint(
start=prog.episode_by_id('breakfast-sequence').start,
end=prog.episode_by_id('choose-beverage-ingredient-ep').start,
ctype='controllable',
lb=0.2,
ub=0.3),
TemporalConstraint(start=prog.episode_by_id('parallel-food-ep').start,
end=prog.episode_by_id('observe-food-ep').start,
ctype='controllable',
lb=0.0,
ub=0.0),
TemporalConstraint(
start=prog.episode_by_id('parallel-food-ep').start,
end=prog.episode_by_id('choose-food-ingredient-ep').start,
ctype='controllable',
lb=0.2,
ub=0.3)
]
for tc in extra_tcs:
prog.add_temporal_constraint(tc)
prog.add_overall_temporal_constraint(ctype='controllable', lb=0.0, ub=7.0)
prog.simplify_temporal_constraints()
return prog
def rmpyl_simple_verbose(hello,uav):
"""Simple RMPyL example using verbose syntax."""
prog = RMPyL()
prog *= prog.sequence(hello.scan(),uav.scan(),prog.parallel(hello.fly(),uav.fly()))
return prog
"""
return Episode(duration={'ctype':'controllable','lb':5,'ub':30},
action='(relay %s)'%(self.name))
loc={'start':(8.751,-8.625),
'minerals':(0.0,-10.0),
'funny_rock':(-5.0,-2.0),
'relay':(0.0,0.0),
'alien_lair':(0.0,10.0)}
rov1 = Rover(name='spirit')
prog = RMPyL(name='run()')#name=run() is a requirement for Enterprise at the moment
prog *= prog.sequence(
rov1.go_to(start=loc['start'],goal=loc['minerals'],risk=0.01),
rov1.go_to(start=loc['minerals'],goal=loc['funny_rock'],risk=0.01),
rov1.go_to(start=loc['funny_rock'],goal=loc['alien_lair'],risk=0.01),
rov1.go_to(start=loc['alien_lair'],goal=loc['relay'],risk=0.01))
tc=prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=2000.0)
cc_time = ChanceConstraint(constraint_scope=[tc],risk=0.1)
prog.add_chance_constraint(cc_time)
#Option to export the RMPyL program to an Enterprise-compliant TPN.
prog.to_ptpn(filename='picard_rovers_rmpyl.tpn')
#Writes RMPyL program to pickle file.
with open('picard_rovers_rmpyl.pickle','wb') as f:
print('Writing RMPyL program to pickle file.')
pickle.dump(prog,f)
domain,problem,task = model_parser(dom_file,prob_file,remove_static=True)
start = time.time()
sat_plans = py_sat.plan(task.initial_state,task.goals,time_steps=18)
elapsed = time.time()-start
print('\n##### All solving took %.4f s'%(elapsed))
if len(sat_plans)>0:
plan = sat_plans[0]
print('\n##### Plan found!\n')
for t,action in enumerate(plan):
print('%d: %s'%(t,action))
prog = RMPyL(name='run()')
prog.plan = prog.sequence(*[Episode(start=Event(name='start-of-'+op),
end=Event(name='end-of-'+op),
action=op,
duration=rss_duration_func(op)) for op in plan])
prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=2000.0)
prog.to_ptpn(filename='rss_pysat_before_stnu_reform.tpn')
paris = PARIS()
risk_bound,sc_sched = paris.stnu_reformulation(prog,makespan=True,cc=0.001)
if risk_bound != None:
risk_bound = min(risk_bound,1.0)
print('\nSuccessfully performed STNU reformulation with scheduling risk %f %%!'%(risk_bound*100.0))
prog.to_ptpn(filename='rss_pysat_after_stnu_reform.tpn')
print('\nThis is the schedule:')
for e,t in sorted([(e,t) for e,t in sc_sched.items()],key=lambda x: x[1]):
print('\t%s: %.2f s'%(e,t))
else:
# print("---------plan: %d" % len(sat_plans))
if len(sat_plans)>0:
plan = sat_plans[0] # get the first plan (default returns a list with one plan)
for t,op in enumerate(plan):
print('%d: %s'%(t,op))
elapsed = time.time()-start
print('\n##### All solving took %.4f s'%(elapsed))
prog = RMPyL(name='run()')
pddl_episodes = [Episode(id=make_episode_id(t,op),
start=Event(name='start-of-%d-%s'%(t,op)),
end=Event(name='end-of-%d-%s'%(t,op)),
action=op,
duration=rss_duration_model_func(op)) for t,op in enumerate(plan)]
prog.plan = prog.sequence(*pddl_episodes)
# prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=2000.0)
#Adds temporal window to the plan
for t,op in enumerate(plan):
bounds, tc_type = rss_time_window_model_func(op)
for tc in time_window_constraints(tc_type,bounds,prog.first_event,prog.episode_by_id(make_episode_id(t,op))):
prog.add_temporal_constraint(tc)
#Dummy episodes that enable transmissions
activation_episodes=[]
activation_tcs=[]
global_start=Event(name='global-start')
for op_name,op_param_dict in time_windows['time_windows'].items():
for arg_set,window_dict in op_param_dict.items():
for ev_type,time_bound in window_dict.items():
verbose=1)
#Searches for the optimal policy
policy, explicit, performance = planner.search(b0)
#Converts policy to graphical SVG format
dot_policy = policy_to_dot(explicit, policy)
dot_policy.write('flightgear_policy.svg', format='svg')
#Converts optimal exploration policy into an RMPyL program
exploration_policy = policy_to_rmpyl(explicit, policy)
#The flight policy has the additional actions of taking off and landing.
flight_policy = RMPyL(name='run()')
flight_policy *= flight_policy.sequence(Episode(action='(takeoff plane)'),
exploration_policy,
Episode(action='(land plane)'))
#Eliminates probabilistic choices from the policy, since Pike (in fact, the
#Lisp tpn package) cannot properly handle them.
for obs in flight_policy.observations:
if obs.type == 'probabilistic':
obs.type = 'uncontrollable'
del obs.properties['probability']
#Converts the program to a TPN
flight_policy.to_ptpn(filename='flightgear_rmpyl.tpn')
# Writes control program to pickle file
with open('flightgear_rmpyl.pickle', 'wb') as f:
pickle.dump(flight_policy, f, protocol=2)
halt_on_violation=False,verbose=1)
#Searches for the optimal policy
policy,explicit,performance = planner.search(b0)
#Converts policy to graphical SVG format
dot_policy = policy_to_dot(explicit,policy)
dot_policy.write('flightgear_policy.svg',format='svg')
#Converts optimal exploration policy into an RMPyL program
exploration_policy = policy_to_rmpyl(explicit,policy)
#The flight policy has the additional actions of taking off and landing.
flight_policy = RMPyL(name='run()')
flight_policy *= flight_policy.sequence(Episode(action='(takeoff plane)'),
exploration_policy,
Episode(action='(land plane)'))
#Eliminates probabilistic choices from the policy, since Pike (in fact, the
#Lisp tpn package) cannot properly handle them.
for obs in flight_policy.observations:
if obs.type=='probabilistic':
obs.type = 'uncontrollable'
del obs.properties['probability']
#Converts the program to a TPN
flight_policy.to_ptpn(filename='flightgear_rmpyl.tpn')
# Writes control program to pickle file
with open('flightgear_rmpyl.pickle','wb') as f:
pickle.dump(flight_policy,f,protocol=2)
def rmpyl_icaps14():
"""
Example from (Santana & Williams, ICAPS14).
"""
prog = RMPyL()
prog *= prog.decide(
{
'name': 'transport-choice',
'domain': ['Bike', 'Car', 'Stay'],
'utility': [100, 70, 0]
},
prog.observe(
{
'name': 'slip',
'domain': [True, False],
'ctype': 'probabilistic',
'probability': [0.051, 1.0 - 0.051]
},
prog.sequence(
Episode(action='(ride-bike)',
duration={
'ctype': 'controllable',
'lb': 15,
'ub': 25
}),
Episode(action='(change)',
duration={
'ctype': 'controllable',
'lb': 20,
'ub': 30
})),
Episode(action='(ride-bike)',
duration={
'ctype': 'controllable',
'lb': 15,
'ub': 25
})),
prog.observe(
{
'name': 'accident',
'domain': [True, False],
'ctype': 'probabilistic',
'probability': [0.013, 1.0 - 0.013]
},
prog.sequence(
Episode(action='(tow-vehicle)',
duration={
'ctype': 'controllable',
'lb': 30,
'ub': 90
}),
Episode(action='(cab-ride)',
duration={
'ctype': 'controllable',
'lb': 10,
'ub': 20
})),
Episode(action='(drive)',
duration={
'ctype': 'controllable',
'lb': 10,
'ub': 20
})), Episode(action='(stay)'))
prog.add_overall_temporal_constraint(ctype='controllable', lb=0.0, ub=30.0)
return prog