18
The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology {bclement, mdj}@aig.jpl.nasa.gov http://ai.jpl.nasa.gov/

The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Embed Size (px)

Citation preview

Page 1: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

The Deep Space Network Scheduling Problem

The Deep Space Network Scheduling Problem

Brad Clement, Mark Johnston

Artificial Intelligence Group

Jet Propulsion Laboratory

California Institute of Technology{bclement, mdj}@aig.jpl.nasa.gov

http://ai.jpl.nasa.gov/

Brad Clement, Mark Johnston

Artificial Intelligence Group

Jet Propulsion Laboratory

California Institute of Technology{bclement, mdj}@aig.jpl.nasa.gov

http://ai.jpl.nasa.gov/

Page 2: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Space Networks

Page 3: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Deep Space Network (DSN)

Page 4: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Activities, Tracks, & Viewperiods

• A track is an allocation of an antenna to a mission over some time interval

• A viewperiod is the time interval when a spacecraft is visible to an antenna

• An activity is a track wrapped with setup and teardown time

Page 5: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Deep Space Network Scheduling

• 56 missions

• 12 antennas– different capabilities

– shared equipment

– geometric constraints

– human operator constraints

• some schedule as long as 10 years into future

• ~370 tracks & ~1650 viewperiods per week

• ~2000 tracks & ~80000 viewperiods per year

• some require schedule freeze 6 months out

• complicated requirements originally from agreement with NASA with flexibility in antennas, timing, numbers of tracks, gaps, etc.

• ~30 people employed full time to schedule for multiple missions

• schedule centrally generated, meetings and horse trading to resolve conflicts

• similar to coordination operations across missions

Page 6: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Other Space Scheduling Problems

• comm. scheduling TDRSS and AFSCN

• Mars relay scheduling

• antenna command generation

• science planning

• measurement scheduling

• command sequence generation (ground and onboard)

Page 7: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Constraints• No two spacecraft can use

antenna at same time– except MSPA where antenna

points to both (2 at most) and uplinks to at most one

• Spacecraft must be in view of antenna

• At Goldstone, no track/activity can be scheduled where two other tracks/activities start within 15 minutes– except the four Cluster s/c

• At other complexes, no two may start within 5 minutes of each other

U/LD/L

D/L U/L

D/L

MEXMERA

NAV

0

5

2 3 2 232

30

Page 8: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Requirements

– abstraction of requirements as an AND/OR tree→use HTN planning

– optional and/or multiple resource usage – start time and duration ranges– temporal constraints (STN)– all activity/track start times and

durations must be evenly divisible by 5 minutes (except for Cluster)

– locks on resource and timing→ remove resource choices (OR branches)→add/shrink temporal constraints

to current time allocation→ASPEN has scheduling permissions

– override / blockout

[3hr, 8hr]

0

0

Page 9: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Requirements (cont’d)

periodic tracks – ranges specifiedfor some of:

• initial start time• overall end of period• number of tracks• total duration of tracks• duration of individual

tracks• time gap/overlap

gap[min,max]

d1 d2

consumptionequal toduration

cons

umab

lere

sour

ce min total duration(added at end)max total

duration must scheduledi to not overfillor exhaust

Page 10: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Scheduling in ASPEN

Start (if conflicts exist and user time-limit not exceeded)

...Select a conflict

Select a repair method...

move

...

...

Select an activity

Select a start time

Perform theaction, collectthe new conflicts,and repeat

Page 11: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Scheduling Performance

• Generates schedule of 1861 tracks from ~3 weeks of requests in 39 minutes (resolving 2305 initial view period/antenna conflicts)

• Reschedules to accommodate individual emergency tracks in 0.2 seconds and emergency antenna downtime in 0.2 seconds

• Handles doubling of one mission’s track requests over one week (to 42 total) in 2.7 seconds

• Initial performance acceptable for interactive conflict resolution, possibly for initial schedule generation

Page 12: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Systematic Search Algorithms• Local search in ASPEN can handle large schedules (for

long-term requirements) but gives no guarantees of optimality or that an existing solution will be found

• Systematic search can give these guarantees for small problems (conflict resolution)– BT1: depth-first backtracking, all times/resources permitted

(provided there are viewperiods available)• most constrained first track selection• smallest from original time/antenna assignment

– BT2: same as BT1 but limited to original antenna– A*: optimal graph search, objective is to minimize changes from

original schedule (in both both antenna and time)– Assumptions made for these algorithms and experiments:

• track durations are fixed in these experiments• tracks that span day boundaries are considered locked in place• tracks without viewperiods are considered locked in place

Page 13: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

M. Johnston

Resolving conflicts forSchedule after Conflict Negotiation Meeting

Finds solution (or proves no solution) to each within a few hundredths of a second

Found optimal solutions to most within a few minutes

Page 14: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

M. Johnston

Finds solution to each within a few hundredths of a second

Found optimal solutions to most within a few minutes

Resolving conflicts Schedule Before Conflict

Negotiation Meeting

Page 15: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

User-Interface Design

• Met repeatedly with users (MDAPT, RAPSO, DSN, and mission operations staff) to understand requirements and obtain feedback on application design.

• Collaboration with NASA Ames’ Human Computer Interaction group (Alonzo Vera, Mike McCurdy & Chris Connors), who (with us and user feedback) have designed user-interfaces.

– designed interface forediting of requirements

– designed interface forrefining requirements withaid of automated scheduler

Page 16: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Requirements Editor Dialog

Page 17: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Visualization

• Gantt chart design• Mouse over details in

tabular view• simultaneous schedule +

metric visibility w/user-specifiable gradient

• visualization of differences between two schedule versions

Page 18: The Deep Space Network Scheduling Problem Brad Clement, Mark Johnston Artificial Intelligence Group Jet Propulsion Laboratory California Institute of Technology

Applications – DSN Arrays

• NASA may build 3600 10m weather-sensitive antennas

• 1200 at each complex in groups of 100 spread over wide area

• High automation requested—one operator for 100 or 1200 antennas

• Spacecraft may use any number of antennas for varying QoS, and may need link carried across complexes

• Only some subsets of antenna signals can be combined

– depends on design of wiring/switching to combiners

– combiners may be limited

• Local response time should be minimized

DSCC

Array Signal Proc

Other DSN Systems

Array Sites

Sig ProcSig Proc

Sig Proc

Sig Proc

Sig Proc

Sig Proc

Sig Proc

Sig Proc