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 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_parallel_choices(hello,uav):
"""Simple RMPyL example with parallel execution of choices."""
uav2 = UAV(name='uav2')
prog = RMPyL()
prog *= prog.parallel(
prog.observe({'name':'HELLO-OBS','domain':['FLY','SCAN','CRASH'],
'ctype':'probabilistic','probability':[0.50,0.49,0.01]},
hello.fly(),hello.scan(),hello.crash()),
prog.observe({'name':'UAV-OBS','domain':['FLY','SCAN','CRASH'],
'ctype':'probabilistic','probability':[0.50,0.49,0.01]},
uav.fly(),uav.scan(),uav.crash()))*uav2.fly()
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_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
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():
orb_ep_id='%s_event_%s-%s'%(ev_type,op_name,'-'.join(arg_set))
activation_episodes.append(Episode(id=orb_ep_id,
action='(%s)'%(orb_ep_id.replace('-',' ')),
duration={'ctype':'controllable','lb':0.005,'ub':0.1}))
activation_tcs.append(TemporalConstraint(start=global_start,
end=activation_episodes[-1].start,
ctype='controllable',
lb=float(time_bound),
ub=float(time_bound)))
global_prog = RMPyL(name='run()')
global_prog *= global_prog.parallel(prog.plan,*activation_episodes,start=global_start)
activation_tcs.append(TemporalConstraint(start=global_start,end=prog.first_event,
ctype='controllable',lb=0.0,ub=0.005))
for tc in activation_tcs:
global_prog.add_temporal_constraint(tc)
# rmpyl_policy.to_ptpn(filename=output_file)
#ipdb.set_trace()
global_prog.to_ptpn(filename=output_file+'_test')
paris = PARIS()
# risk_bound,sc_schedule = paris.stnu_reformulation(rmpyl_policy,makespan=True,cc=0.001)
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
action='(%s)' % (orb_ep_id.replace('-', ' ')),
duration={
'ctype': 'controllable',
'lb': 0.005,
'ub': 0.1
}))
activation_tcs.append(
TemporalConstraint(start=global_start,
end=activation_episodes[-1].start,
ctype='controllable',
lb=float(time_bound),
ub=float(time_bound)))
global_prog = RMPyL(name='run()')
global_prog *= global_prog.parallel(prog.plan,
*activation_episodes,
start=global_start)
activation_tcs.append(
TemporalConstraint(start=global_start,
end=prog.first_event,
ctype='controllable',
lb=0.0,
ub=0.005))
for tc in activation_tcs:
global_prog.add_temporal_constraint(tc)
# rmpyl_policy.to_ptpn(filename=output_file)
#ipdb.set_trace()