Ames Research Center Incremental Contingency Planning Richard Dearden, Nicolas Meuleau, Sailesh Ramakrishnan, David E. Smith, Rich Washington window [10 ,14:30] power power Workspace pan data Drive (- 1) Dig(5) Visual servo (.2, -.15) NIR Lo res Rock finder Hi res Carbonate [10 ,14:30] X X X X ?
40
Embed
Ames Research Center Incremental Contingency Planning Richard Dearden, Nicolas Meuleau, Sailesh Ramakrishnan, David E. Smith, Rich Washington window [10,14:30]
Ames Research Center The Planning Problem Visual servo (.2, -.15) Warmup NIR Dig(5)Drive(-1)NIR ……… Compress Drive(2) Maximize (Expected) Scientific Return Given: start time pose energy available actions with uncertain: durations resource usage Possible science objectives images samples
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
AmesResearchCenter
Incremental Contingency Planning Richard Dearden, Nicolas Meuleau,
Sailesh Ramakrishnan, David E. Smith, Rich Washington
window
[10 ,14:30]
power power
Workspace pandata
Drive (-1)Dig(5)Visual servo (.2, -.15) NIR
Lo res Rock finder Hi res Carbonate
[10 ,14:30]
X X XX?
AmesResearchCenter
Limited onboard processingCPU, memory, time
Safetysequence checking
Anticipationsetup steps
Why Contingency Planning ??
AmesResearchCenter The Planning Problem
Visual servo (.2, -.15)
Warmup NIR
Dig(5) Drive(-1) NIR ………Compress
Drive(2)
Maximize (Expected) Scientific Return
Given:start timeposeenergy availableactions with uncertain:
durationsresource usage
Possible science objectivesimagessamples
AmesResearchCenter Technical Challenges
Continuous time (& resources)
Continuous outcomes
Time (& resource) constraints
Concurrency
Goal selection & optimizationg1, g2, g3, g4 …
Visual servo (.2, -.15)
Warmup NIR
Lo res Rock finder NIR
∆p =∆t =
NIR
E > 2 Aht [10:00, 14:00]
Time Power Storage
AmesResearchCenter Just in Case (JIC) Scheduling
1. Seed schedule2. Identify most likely failure3. Generate a contingency branch4. Integrate the branch
Advantages: TractabilitySimple plansAnytime
.4 .2.1
AmesResearchCenter Just in Case (JIC) Planning
1. Seed plan2. Identify most likely failure3. Generate a contingency branch4. Integrate the branch
.4 .2.1
AmesResearchCenter Limits of JIC Scheduling Heuristics
Dig(60)Visual servo (.2, -.15)
Lo res Rock finder LIB
= 120s = 60s
= 300s = 5s
= 1000s = 500s
t [9:00, 16:00] = 5s = 1s
= 120s = 20s V = 50
HiRes
V = 10
t [10:00, 13:50] = 600s = 60s
t [9:00, 14:30] = 5s = 1s
V = 5
Warmup LIB
= 1200s = 20s
Most probable failure points may not be the best branch-points:
It is often too late to attempt other goals when the plan is about to fail.
: most probable failures$ : most interesting branch point
ExpectedUtility
PowerStart time
10
1520
5
13:20
14:4014:20
14:0013:40
True for all initial states in the grey box.
Drive (-2) NIR
V = 100
t [10:00, 14:00] = 600s = 60s $
AmesResearchCenter Just in Case (JIC) Planning
1. Seed plan2. Identify best branch point3. Generate a contingency branch5. Evaluate & Integrate the branch
? ??
Construct plangraph
Back-propagate value tables
Compute gain
Select branch condition & goals
?
r
Vb
Vm
AmesResearchCenter Construct Plangraph
g1
g2
g3
g4
AmesResearchCenter Value Tables
g1
g2
g3
g4
V1
V2
V3
V4
r
r
r
r
AmesResearchCenter Example
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
AmesResearchCenter Simple Propagation
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
AmesResearchCenter Simple Propagation
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
AmesResearchCenter Simple Propagation
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
12
AmesResearchCenter Conjunctions
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
12
12
r
12
t
AmesResearchCenter Propagation
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
12
12
r
12
t
51
r
AmesResearchCenter Propagation
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
12
12
r
12
t
51
r
15
t
115
t
AmesResearchCenter Combining Tables
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
12
12
r
12
t
51
r
15
t
115
t
15
t
AmesResearchCenter Discharging Assumptions
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
12
r
12
t
51
r
15
t
115
t
15
t
15
15
AmesResearchCenter Propagation
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
5
25
15
15
16
AmesResearchCenter Combining Tables
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
15
15
61
18
5
2515
AmesResearchCenter Combining Tables
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
15
15
61
5
2515
5
258
18
AmesResearchCenter Ordering
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
15
15
61
5
2515
5
258
DCE
CDE
AB
18
AmesResearchCenter Achieving Multiple Goals
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
15
15
61
5
2515
5
258
18
30
g+g’g
g’ +
AmesResearchCenter Achieving Multiple Goals
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
15
15
61
18
5
2515 30
g+g’g
g’
5g+g’
g
g’
258 30
AmesResearchCenter Goal Annotation
A
E
B
e
5
(1, 5)
(3, 3)
(10, 15) (10, 15)
(2, 2)
C
Ds t
r
q
p
g
g’
e1
5
15
15
15
61
18
gg
g’
g’
g’g’
g’
5
2515 30
g+g’g
g’
5g+g’
g
g’
258 30
AmesResearchCenter Just in Case (JIC) Planning
1. Seed plan2. Identify best branch point3. Generate a contingency branch5. Evaluate & Integrate the branch
? ??
Construct plangraph
Back-propagate value tables
Compute gain
Select branch condition & goals
?
r
Vb
Vm
AmesResearchCenter Estimating Branch Value
V1
V2
V3
V4
V
r
V
r
V
r
MaxV
r
AmesResearchCenter Plan Statistics
r
V1
V2
V3
V4
P
r
plan value functionresource probability
Vm
Vb
r
AmesResearchCenter Expected Branch Gain
V1
V2
V3
V4
P
r
Gain = ∫ P(r) max{0,Vb(r) - Vm(r)} dr∞
0
Vb
r
rVb
Vm
AmesResearchCenter Selecting the Branch Condition
V1
V2
V3
V4
P
r
branch condition
rVb
Vm
Vb
r
branch condition
AmesResearchCenter Selecting Branch Goals
r
V1
V2
V3
V4
P
r
branch goals
g1
g3
g3
g1
Vb
r
rVb
Vm
AmesResearchCenter Evaluating the Branch
1. Seed plan2. Identify best branch point3. Generate a contingency branch4. Evaluate & integrate the branch
? ?? ?
r
Vb
Vm Compute value function
Compute actual gain
AmesResearchCenter Actual Branch Gain
rVb
P
r
Gain = ∫ P(r) max{0,Vb(r) - Vm(r)} dr∞
0
Vm
r
Vb Branch value function
actual branch condition
AmesResearchCenter Remarks: Single Plangraph
1. Seed plan2. Identify best branch point3. Generate a contingency branch5. Evaluate & Integrate the branch
? ??
Construct plangraph
Back-propagate value tables
Compute gain
Select branch condition & goals
?
r
Vb
Vm
AmesResearchCenter Plan Graph
g1
g2
g3
g4
V1
V2
V3
V4
r
r
r
r
AmesResearchCenter Branch Initial Conditions
g1
g2
g3
g4
V1
V2
V3
V4
r
r
r
r
v
rv
r
v
rv
r
v
r
{p}
{q,r}
{p,r}
AmesResearchCenter Single Plangraph
1. Seed plan2. Identify best branch point3. Generate a contingency branch4. Evaluate & integrate the branch