jonne zutt delft university of technology information technology and systems
DESCRIPTION
TRAIL/TNO Project 16. Fault detection and recovery in multi-modal transportation networks with autonomous mobile actors. Jonne Zutt Delft University of Technology Information Technology and Systems Collective Agent Based Systems Group. Supervisors Dr. C. Witteveen Dr. ir. Z. Papp - PowerPoint PPT PresentationTRANSCRIPT
12/9/04 – Review TNO/TRAIL project #16 1
Jonne ZuttDelft University of Technology
Information Technology and Systems
Collective Agent Based Systems Group
Fault detection and recovery in multi-modaltransportation networks with autonomous mobile actors
TRAIL/TNO Project 16
Supervisors
Dr. C. Witteveen
Dr. ir. Z. Papp
Dr. ir. A.J.C. van Gemund
12/9/04 – Review TNO/TRAIL project #16 2
Contents
• Transportation planning• Problem description• Progress• Methods and hypotheses• Experiments
12/9/04 – Review TNO/TRAIL project #16 3
Issues in design and control of MHS
• Guide-path design• Estimating optimal
number of vehicles• Vehicle maintenance• Order allocation• Idle-vehicle positioning• Vehicle routing• Conflict-resolution
12/9/04 – Review TNO/TRAIL project #16 4
Layers
• Guide-path design• Estimating optimal
number of vehicles• Vehicle maintenance• Order allocation• Idle-vehicle positioning• Vehicle routing• Conflict-resolution
Strategic
Tactic
Operationalminutes
hours
months
12/9/04 – Review TNO/TRAIL project #16 5
Problem description
• Design a model for operational transport planning,
• Develop multi-agent routing and scheduling methods that can take into account incidents,
• Search suitable performance indicators to be used in experiments for comparing the quality of different methods taking into account properties of the environment.
12/9/04 – Review TNO/TRAIL project #16 6
Progress – previous years
• Model for operational transport planning
• Methods for operational transport planning taking into account incidents
• Transport planning simulator
12/9/04 – Review TNO/TRAIL project #16 7
Progress – last year
• Test set• Performance indicators• Experimental results• Thesis structure• Approximately two chapters written
12/9/04 – Review TNO/TRAIL project #16 8
Progress – future work
• Complete single-agent experiments [December’04]
• Coordination experiments [February’05]
• Writing [June’05]
12/9/04 – Review TNO/TRAIL project #16 9
Overview methods
fixed routing
rerouting
Arb-ci
HNZ-0HN
LPA*HNZ
no planning look-ahead
strictcommitments
loose commitments/ decommitments
hi
bj
hi b
j r
k
12/9/04 – Review TNO/TRAIL project #16 10
Conflicts1. Resources have limited capacity
A B C
2. Instantaneous exchange
ABDTime
A B C
ABTime
D
12/9/04 – Review TNO/TRAIL project #16 11
About cycles and deadlocks
A
B C
K(A)=1
P(K_sema_C)V(K_sema_B)
A
B
History: F,E,D,CCurrent: B,A
12/9/04 – Review TNO/TRAIL project #16 12
Methods – Simple/plan-based arbiter policies
• First-In-First-Out• Agent priority• Longest-Queue-First• Longest-Queue-First-Inc
• Longest-Plan-First• Most-Urgent-Deadline-First• Max-Reward-Decrease-First• Max-Reward-Decrease-Queue-First
Hypothesis:
No/very small
difference Hypothesis:
Plan-based policies
outperform the simple policies
12/9/04 – Review TNO/TRAIL project #16 13
Methods – HNZ
• Wait for a change in plan(s)• While agents are not ready– Compute traffic-aware shortest path– Agent compete who schedules first (P1)–Winner schedules n resources (P2)
• If current order rewards are below threshold, agent tries to reroute (P3)
Hypothesis: Much better than no planning
Hypothesis:Rerouting most important par
12/9/04 – Review TNO/TRAIL project #16 14
Method: agent selection functions (P1)
• RandomProvides a baseline for the others
• DelaysAgent with maximum wait time first
• DeadlinesAgent with most strict deadlines first
• PenaltiesAgent with lowest planned reward first
Hypo: All agent selection functions will outperform random
12/9/04 – Review TNO/TRAIL project #16 15
Method: resource block-size (P2)
• How many resources (fraction of route) are scheduled after the agent is selected by the agent selection function?
Hypothesis:A smaller block-size slightly increases
performance but also increases computation time
12/9/04 – Review TNO/TRAIL project #16 16
Number of reroute opportunities
Number of alternatives
Ave
rage
% o
f del
ay
Number of alternativesTa
rdin
ess
Tardiness aA Ca - a if Ca> aDelay { aA (Ca – Ma) / Ca } / |A|
12/9/04 – Review TNO/TRAIL project #16 17
Agent selectionA
vera
ge su
m o
f del
iver
y pe
nalti
es
No incidents Pfail = 0.1 Pfail = 0.20 reroutes 1 reroute 0 reroutes 1 reroute 0 reroutes 1 reroute
1. Random2. Delays3. Deadlines4. Penalties
0
500
1
000
150
0 2
000
250
0 3
000
350
0
12/9/04 – Review TNO/TRAIL project #16 18
Block size
No incidents Pfail = 0.1 Pfail = 0.2
0 1 1 1 1 1 10 0 0 0 01 1 1 1 1 1
Ave
rage
sum
of d
eliv
ery
pena
lties
2 2 4 6 ∞ 2 4∞ 2 ∞ 2 ∞6 ∞ 2 4 6 ∞
1. max. number of reroutes2. block size
0
100
0
2
000
300
0
12/9/04 – Review TNO/TRAIL project #16 19
Time for different block sizes
No incidents Pfail = 0.1 Pfail = 0.22 2 4 6 ∞ 2 4∞ 2 ∞ 2 ∞6 ∞ 2 4 6 ∞0 1 1 1 1 1 10 0 0 0 01 1 1 1 1 1
Ave
rage
cpu
tim
e
0
1
2
3
4
5
6
71. max. number of reroutes2. block size
12/9/04 – Review TNO/TRAIL project #16 20
Coordination – Coalition Formation
• Static– Different companies
• Dynamic– Based on current position– Based on source/destination locations, or
plan distance function– Grouped orders
Hypothesis:Dynamic coalitions are preferable, though static
coalitions already improve the coalition’s welfare
12/9/04 – Review TNO/TRAIL project #16 21
Coordination – How to improve welfare?
• Exchange orders with coalition members (cf. simulated trading)
• Conflict-resolution:In case of a conflict, determine Δ(C) instead of Δ(A) to determine who wins.
12/9/04 – Review TNO/TRAIL project #16 22
Questions• CABS project:
http://cabs.ewi.tudelft.nl• My homepage: http://dutiih.twi.tudelft.nl
/~jonne• My email:
12/9/04 – Review TNO/TRAIL project #16 23
Thesis1. Introduction
– Challenges in transportation– Problem description– Approach– Research contributions– Overview
2. A model and formalism for multi-agent transport planning
– Introduction– Building blocks– Correctness criteria– Performance criteria
3. Single-agent methods for transport planning
– Order allocation– Operational planning– Route planning– Simple arbiter policies– Revising priorities– Revising route– Lifelong Planning A*
4. Experiments on single-agent methods– Experimental setting– Description of the test set– Experimental results
5. Multi-agent methods for transport planning– Introduction– Coalition formation– Exchanging transportation orders– Conflict solving
6. Experiments on multi-agent methods– Experimental setting– Experimental results
7. Conclusions
A. Mathematical preliminariesB. Complexity of transport planning
12/9/04 – Review TNO/TRAIL project #16 24
Model
Auctioneeragent
Transportagent
Transportagent
Transportagent
Customeragent
Transportresource
Transportresource
Transportresource
speedcapacity
max. speedcapacitydistance
cooperativecompetitive
12/9/04 – Review TNO/TRAIL project #16 25
Model: incidents• Events that disrupt regular plan execution
and generally require re-planning• Examples: customers that change or retract
transportation orders, unpredictable congestion, vehicle break-down, communication failure
• Incidents are generated proportional to the resources. Pfail = 0.x means each resources is expected to fail x·10% of the time.
12/9/04 – Review TNO/TRAIL project #16 26
Method: traffic-aware shortest path
• Agents know which time-windows are in use by other agents per resource
• Run an A* algorithm: store routes on open list, check for conflict when appending to candidate route
• Process is guaranteed to terminate and find the traffic-aware shortest path
12/9/04 – Review TNO/TRAIL project #16 27
Experiments
• 10 transport networks with 25 resources, ‘random’ topology.
• 10 sets of transportation orders with 250 random orders each
• 2 different sets of agents with 25 randomly located agents each
• Incidents with failure probability 0, 0.1, …, 1.0 and impact 0.1.
12/9/04 – Review TNO/TRAIL project #16 28
Blocktime
12/9/04 – Review TNO/TRAIL project #16 29
Simple arbiter policies
12/9/04 – Review TNO/TRAIL project #16 30
HNZ-0/1 150 orders
12/9/04 – Review TNO/TRAIL project #16 31
HNZ-0/1 250 orders