def rmpyl_simple_nested_choice(hello,uav):
"""Simple RMPyL example with nested choice."""
prog = RMPyL()
prog *= prog.decide({'name':'UAV-choice1','domain':['H','U'],'utility':[5,7]},
prog.decide({'name':'UAV-choice2','domain':['H','U'],'utility':[5,7]},
hello.fly(),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 make_fly_scan(hello,uav):
"""Subplan where one choose which UAV should fly and which should scan."""
p_fly_scan = RMPyL()
p_fly_scan*= p_fly_scan.decide({'name':'Choose-FLY','domain':['H','U'],'utility':[5,7]},
hello.fly()+uav.scan(),
hello.scan()+uav.fly())
return p_fly_scan
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_simple_decide_looping(hello,uav):
"""Simple RMPyL example where we choose to fly loops with different UAVs."""
prog = RMPyL()
prog *= prog.decide({'name':'Choose-loop','domain':['H','U'],'utility':[1,1]},
prog.loop(episode_func=hello.fly,repetitions=3,
run_utility=1,stop_utility=0),
prog.loop(episode_func=uav.scan,repetitions=2,
run_utility=2,stop_utility=0))
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 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_parallel_and_choice_user_defined_tcs(hello,uav):
"""Simple RMPyL example with parallel execution of actions on different choice
branches with user-defined constraints."""
prog = RMPyL()
#Choice using the previous example as a subroutine to generate partial progs
prog *= prog.decide({'name':'Choose-branch','domain':['1','2'],'utility':[1,2]},
rmpyl_parallel_user_defined_tcs(hello,uav),
rmpyl_parallel_user_defined_tcs(hello,uav))
prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=70.0)
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 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_choice_risk():
prog = RMPyL()
prog *= prog.decide(
{
'name': 'transport-choice',
'domain': ['Bike', 'Car', 'Stay'],
'utility': [100, 70, 0]
},
prog.observe(
{
'name': 'travel',
'ctype': 'probabilistic',
'domain': ['Short', 'Long'],
'probability': [0.7, 0.3]
},
Episode(action='(cab-ride-long)',
duration={
'ctype': 'uncontrollable_probabilistic',
'distribution': {
'type': 'uniform',
'lb': 5.0,
'ub': 11.0
}
}),
Episode(action='(cab-ride-short)',
duration={
'ctype': 'uncontrollable_probabilistic',
'distribution': {
'type': 'uniform',
'lb': 5.0,
'ub': 10.0
}
})),
Episode(action='(drive-car)',
duration={
'ctype': 'controllable',
'lb': 6.0,
'ub': 8.0
}), Episode(action='(stay)'))
prog.add_overall_temporal_constraint(ctype='controllable', lb=0.0, ub=10.0)
return prog
def rmpyl_choice_risk():
prog = RMPyL()
prog *= prog.decide(
{'name':'transport-choice','domain':['Bike','Car','Stay'],
'utility':[100,70,0]},
prog.observe(
{'name':'travel','ctype':'probabilistic',
'domain':['Short','Long'],'probability':[0.7,0.3]},
Episode(action='(cab-ride-long)',duration={'ctype':'uncontrollable_probabilistic',
'distribution':{'type':'uniform',
'lb':5.0,'ub':11.0}}),
Episode(action='(cab-ride-short)',duration={'ctype':'uncontrollable_probabilistic',
'distribution':{'type':'uniform',
'lb':5.0,'ub':10.0}})),
Episode(action='(drive-car)',
duration={'ctype':'controllable','lb':6.0,'ub':8.0}),
Episode(action='(stay)'))
prog.add_overall_temporal_constraint(ctype='controllable',lb=0.0,ub=10.0)
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_inconsistent_user_defined_tcs(hello,uav):
"""Simple RMPyL example with parallel execution of actions on different choice
branches with user-defined constraints."""
try:
prog = RMPyL()
hello_flights = [hello.fly() for i in range(2)] #two different hello flights
uav_flights = [uav.fly() for i in range(2)] #two different uav flights
#Choice using the previous example as a subroutine to generate partial plans
prog *= prog.decide({'name':'Choose-branch','domain':['1','2'],'utility':[1,2]},
hello_flights[0]*uav_flights[0],
uav_flights[1]*hello_flights[1])
tc = TemporalConstraint(start=hello_flights[0].end,end=hello_flights[1].start,
ctype='controllable',lb=2.0,ub=3.0)
#This is a constraint between disjoint plan branches. Therefore, it MUST
#cause an error of empty constraint support
prog.add_temporal_constraint(tc)
except InconsistentSupportError:
print('Empty support error correctly caught!')
prog=None
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_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_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