Application Example: Motion...
Transcript of Application Example: Motion...
![Page 1: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/1.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16
Application Example: Motion Control Motion Control
• is one of the most important tasks of mobile embedded systems • is subject to real-time and reliability requirements • heavily depends on the sensory input
Single sensors have a partial, inaccurate view Distributed sensor fusion allows achieving a more complete and accurate perception of the environment
LS1 LS2 fused
![Page 2: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/2.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 17
Example: RoboCup
• Getting wide acceptance as standard benchmark for team robotic • Annual world and national championships in robot soccer • Research done as part of a nationwide DFG-program „Cooperating teams of mobile
robots in dynamic environments“
![Page 3: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/3.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 18
Example:RoboCup Where is the ball?
![Page 4: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/4.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 19
Sensor Data Processing – Filters and Abstraction Levels
Motion Control
![Page 5: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/5.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 20
Sensor Data Processing - Multi-Level Sensor Fusion
![Page 6: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/6.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 21
Environment-Dependent Execution Times of the Filter Modules
Execution times depend on input size Execution times depend on input content
![Page 7: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/7.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 22
Environment-Dependent Execution Times of the Filter Modules (cont’d)
execution times of the arc filter
0,00
20,00
40,00
60,00
80,00
100,00
120,00
140,00
160,00
0 50 100 150 200 250
scan no.
exec
utio
n tim
e [m
s]
input for max. execution time
![Page 8: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/8.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 23
Environment-Dependent Execution Times of the Filter Modules (cont’d)
execution times of the arc filter
0,00
20,00
40,00
60,00
80,00
100,00
120,00
140,00
160,00
0 50 100 150 200 250
scan no.
exec
utio
n tim
e [m
s]
input for min. execution time
![Page 9: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/9.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 24
Contour Filter: Execution Time vs. Mean Distance
020000400006000080000100000120000140000160000180000
0 2000 4000 6000 8000 10000
mean distance [mm]
exec
utio
n tim
e [µ
s]
Execution times > 57ms are only observed for distances < 32cm
![Page 10: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/10.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 25
Environment-Dependent Execution Times of the Filter Modules (cont’d)
execution times of the arc filter
Scheduling WCETs is not an acceptable solution to achieve a predictable timing behavior of the filter modules!
0,00
20,00
40,00
60,00
80,00
100,00
120,00
140,00
160,00
0 50 100 150 200 250
scan no.
exec
utio
n tim
e [m
s]
95%-quantile
![Page 11: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/11.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 26
0,00
20,00
40,00
60,00
80,00
100,00
120,00
80 100 120 140
scan no.
time
[ms]
Execution Time
Detection Time
Functional Redundancy in the Arc Filter • The arc filter evaluates potential ball positions • Only the best estimate is relevant to higher layers (corresponds to ball)
ECET
![Page 12: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/12.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 27
Structural Redundancy in the Edge Filter • Aborted instances deliver results for a fraction of the scene • Incomplete results are valuable input for the sensor fusion
![Page 13: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/13.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 28
Structural Redundancy – Distributed Fusion • Environment is observed by several, distributed sensors • Fusion complements missing results from one sensor by results from others
incomplete result of scanner 1 (previous slide)
incomplete result of scanner 2
result of the fusion
![Page 14: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/14.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 29
Application-Level Adaptation • Raising the fusion level allows reducing system load
– Reduced execution times (CPU load) – Reduced data volume (network load)
• Raising the fusion level may decrease accurateness of results
![Page 15: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/15.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 30
Application-Level Adaptation: Accuracy Tradeoff
![Page 16: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/16.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 31
Application-Level Adaptation: Accuracy Tradeoff (cont’d)
![Page 17: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/17.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 32
Application-Level Adaptation: Accuracy Tradeoff (cont’d)
![Page 18: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/18.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 33
Application-Level Adaptation: Accuracy Tradeoff (cont’d)
![Page 19: Application Example: Motion Controlalt.euk.cs.ovgu.de/EuK/lehre/lehrveranstaltungen/ws0809/...Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 16 Application Example: Motion Control](https://reader035.fdocuments.us/reader035/viewer/2022071500/611fa0369a253f36dc2c5851/html5/thumbnails/19.jpg)
Vorlesung “Echtzeitsysteme” WS 08/09 E. Nett 34
Exploiting Inherent Redundancy • To tolerate transient faults (short term peaks), use application-inherent redundancy
– Functional redundancy within the module instances
Design modules as any-time algorithms
– Structural redundancy within the executions of the sensor fusion
Combine results from distributed sensors
– Time redundancy within a sequence of executions
Schedule a frequency above the minimum
• To tolerate permanent faults (persistent overload), use application-level adaptation (graceful degradation)
– TAFT supports detection of persistent overload
– Application-level adaptation allows lowering system load