![Page 1: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/1.jpg)
A Content-Centric Network for Autonomous Driving
Swarun Kumar
Lixin Shi, Stephanie Gil, Nabeel Ahmed,
Dina Katabi and Daniela Rus
![Page 2: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/2.jpg)
Much Interest in Autonomous Vehicles
DARPA Urban Challenge (Team MIT)
Google’s Autonomous Car
Robots for Disaster Areas (Fukushima)
• Projected to save $100B/yr in US alone [WPI’07]
• Benefits include lower traffic congestion, higher fuel efficiency, improved productivity
![Page 3: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/3.jpg)
Much Interest in Autonomous Vehicles
DARPA Urban Challenge (Team MIT)
Google’s Autonomous Car
Robots for Disaster Areas (Fukushima)
• Projected to save $100B/yr in US alone [WPI’07]
• Benefits include lower traffic congestion, higher fuel efficiency, improved productivity
![Page 4: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/4.jpg)
• Nevada legalized testing autonomous vehicles. California, Florida expected to follow.
• Autonomous Vehicles tested on Europe’s roads
“Expect them on the road by 2020”- General Motors
![Page 5: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/5.jpg)
Challenge: Safely Detecting Hidden Objects
• Sensors on a car see only line of sight objects
![Page 6: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/6.jpg)
Challenge: Safely Detecting Hidden Objects
• Sensors on a car see only line of sight objects
• Hidden objects affect autonomous cars
– “Google’s autonomous car requires occasional human intervention to prevent accident”
– “Future of autonomous driving depends on detecting hidden objects & blind spots” - DARPA Challenge [JFR08]
![Page 7: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/7.jpg)
How can autonomous vehicles detect hidden objects?
Communication
![Page 8: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/8.jpg)
How can autonomous vehicles detect hidden objects?
Communication
• Expand field of view beyond line of sight • Also valuable for human drivers – can
react faster to objects they couldn’t see
![Page 9: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/9.jpg)
Simply use past work on VANETs?
VANETs typically oblivious to application
–Efficient routing
–Reliable message delivery
–… (etc)
But, not integrated with specific applications
![Page 10: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/10.jpg)
Autonomous Driving needs Tight Integration with Communication
• Data is huge and time critical
Communication should focus on information most critical to the application
![Page 11: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/11.jpg)
• Don’t know who has the desired content
– In typical networks, you know your destination
– Instead, autonomous car seeks sensor data from part of the road, e.g. an intersection
– It doesn’t know which car has this information
– Focus on content as opposed to accessing a particular destination
Autonomous Driving needs Tight Integration with Communication
• Data is huge and time critical
Communication should focus on information most critical to the application
![Page 12: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/12.jpg)
• Integrates communication with path planning and navigation in autonomous vehicles
• Has a content centric design
– content, i.e. parts of road, is a first class citizen
• New MAC design where content, not senders, contend for the medium
• Implemented & evaluated on real autonomous vehicles
CarSpeak
![Page 13: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/13.jpg)
1. What is “content” and how do we represent it?
2. How do we disseminate this content?
![Page 14: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/14.jpg)
1. What is “content” and how do we represent it?
2. How do we disseminate this content?
![Page 15: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/15.jpg)
What is “content” and how do we represent it?
• Content is sensor data from cubes in the environment
How do we represent these cubes in environment?
![Page 16: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/16.jpg)
• Zoom in for higher resolution view of a smaller part of environment
Ideally…
• Obtain low resolution view of environment
![Page 17: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/17.jpg)
• Zoom in for higher resolution view of a smaller part of environment
Ideally…
• Obtain low resolution view of environment
![Page 18: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/18.jpg)
• Zoom in for higher resolution view of a smaller part of environment
Ideally…
• Obtain low resolution view of environment
![Page 19: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/19.jpg)
• Zoom in for higher resolution view of a smaller part of environment
Ideally…
• Obtain low resolution view of environment
Need recursive representation that makes best use of available wireless bandwidth
![Page 20: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/20.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
![Page 21: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/21.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
![Page 22: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/22.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
![Page 23: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/23.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
![Page 24: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/24.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
![Page 25: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/25.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
![Page 26: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/26.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
CubesTree Representation
![Page 27: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/27.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
CubesTree Representation
Vertex ID
![Page 28: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/28.jpg)
Representing Content in CarSpeak• Consider large cube encompassing environment
• Recursively divide into 8 smaller cubes
• Car needs cube at resolution
CubesTree Representation
Depth
Vertex ID
, depth)(vertex ID
![Page 29: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/29.jpg)
What Info does Autonomous Car Need?
• Looks for obstacle free paths to destination
• Needs to know which parts of environment:
– Are empty and safe to pass through
– Are occupied and unsafe to pass through
X
![Page 30: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/30.jpg)
What Info does Autonomous Car Need?
• Each cube has one bit: Empty (0) or Occupied (1)
• If cube is empty
all cubes inside are empty
0
0 0
0 0
0 0
0
![Page 31: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/31.jpg)
What Info does Autonomous Car Need?
• Each cube has one bit: Empty (0) or Occupied (1)
• If cube is empty
all cubes inside are empty
• If cube is occupied
at least one cube inside is occupied
0 1
0 0
1
![Page 32: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/32.jpg)
Compactly Representing Data
• Level 1 has 8 bits where 0-empty, 1-occupied
1
0 0 0 1 0 0 0 0
![Page 33: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/33.jpg)
Compactly Representing Data
• Level 1 has 8 bits where 0-empty, 1-occupied
• None of 0 nodes need to be expanded
1
0 0 0 1 0 0 0 0
![Page 34: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/34.jpg)
Compactly Representing Data
• Level 1 has 8 bits where 0-empty, 1-occupied
• None of 0 nodes need to be expanded
• Expand 1 node to see inside at more resolution
1
0 0 0 1 0 0 0 0
0 0 1 0 0 0 0 0
![Page 35: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/35.jpg)
Compactly Representing Data
• Level 1 has 8 bits where 0-empty, 1-occupied
• None of 0 nodes need to be expanded
• Expand 1 node to see inside at more resolution
1
0 0 0 1 0 0 0 0
0 0 1 0 0 0 0 0
0 0 1 0 0 0 0 1
![Page 36: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/36.jpg)
Compactly Representing Data
• Level 1 has 8 bits where 0-empty, 1-occupied
• None of 0 nodes need to be expanded
• Expand 1 node to see inside at more resolution
1
0 0 0 1 0 0 0 0
0 0 1 0 0 0 0 0
0 0 1 0 0 0 0 1Tree representation compresses data efficiently
![Page 37: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/37.jpg)
1. What is “content” and how do we represent it?
2. How do we disseminate this content?
![Page 38: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/38.jpg)
How do we disseminate this content?
Autonomous cars collect huge amount of data
Cannot flood medium with all their data
![Page 39: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/39.jpg)
A Request-Response Approach
![Page 40: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/40.jpg)
A Request-Response Approach
Car requests only data of interest
◦ E.g. at blind spots, intersections, etc.
Requestfor R
Response for R
Cube R
Cars which sense the data, may respond
![Page 41: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/41.jpg)
Challenge 1: Who Should Respond?
![Page 42: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/42.jpg)
Challenge 1: Who Should Respond?
Naïve solution 1: Simply let all cars respond
A lot of redundant data
Cube R
![Page 43: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/43.jpg)
Challenge 1: Who Should Respond?
Naïve solution 2: One car respond; others who hear it suppress their response
Responder leaves before sending all packets
Different cars have different perspectives
Cube R
![Page 44: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/44.jpg)
Challenge 1: Who Should Respond?
Naïve solution 2: One car respond; others who hear it suppress their response
Responder leaves before sending all packets
Different cars have different perspectives
Cube RNeed to balance data diversity with data overlap
![Page 45: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/45.jpg)
Solution: Random Walks
Content of the cube (i.e., its subtree) is divided into packets
Each car uses a different random walk to transmit packets
If one car transmits Eventually finishes walk
If multiple cars transmit Overlap is minimum
![Page 46: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/46.jpg)
Cube 1Cube 2
Challenge 2: Ensure medium is shared fairly by requested content
Request for Cube 1 Request for Cube 2
![Page 47: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/47.jpg)
Challenge 2: Ensure medium is shared fairly by requested content
Cube 1Cube 2
5 cars see Cube 1 One car sees Cube 2
Request for Cube 1 Request for Cube 2
![Page 48: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/48.jpg)
Challenge 2: Ensure medium is shared fairly by requested content
802.11 shares medium between senders Cube 1’s share = 5 x Cube 2’s share
Ideally, we want a MAC where
Cube 1 share = Cube 2’s share
Cube 1Cube 2
![Page 49: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/49.jpg)
Cube 1Cube 2
Solution: Replace sender-contention by content-contention
![Page 50: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/50.jpg)
Instead of senders, cubes contend for the medium Requested cubes get equal share of medium
But cubes are virtual entities Cars viewing a cube contend on its behalf
Cube 1Cube 2
Solution 2: Replace sender-contention by content-contention
![Page 51: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/51.jpg)
Content-Based Contention
C 1 C 2
Each cube should get a share of 1/2
Green car share of the medium 3/4
red car share of the medium 1/4
![Page 52: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/52.jpg)
Content-Based Contention
C 1 C 2
Car listens to how many cars respond for a particular cube
But how can a car compute its medium share?
Share-per-cube = 1 / # cars responding
Car’s total share = (Σ share-per-cube) / # requested cubes
Set contention window using car’s share on off-the-shelf cards
![Page 53: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/53.jpg)
Empirical Results
![Page 54: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/54.jpg)
CarSpeak Implementation
Implemented in Robot OS (ROS)
Integrated with MIT’s Path Planner from DARPA challenge
MAC implemented in the ath9k driver for Atheros WiFi cards
![Page 55: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/55.jpg)
Compared Schemes
CarSpeak
802.11 – Request & Response
Experiments
Indoor Testbed with Robots
Outdoor Testbed with Autonomous Car
![Page 56: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/56.jpg)
Indoor Testbed 10 Roomba robots with Kinect
Navigate environment with obstacles
![Page 57: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/57.jpg)
Can CarSpeak’s MAC assign fair share to content?
2 requested cubes, 10 moving robots
0
0,2
0,4
0,6
0,8
1
0 1 2 3 4
Ratio of data transmitted from Cube 1 to Cube 2
CD
F Ideal System
![Page 58: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/58.jpg)
Can CarSpeak’s MAC assign fair share to content?
2 requested cubes, 10 moving robots
0
0,2
0,4
0,6
0,8
1
0 1 2 3 4
Ratio of data transmitted from Cube 1 to Cube 2
802.11 – Req&Res
CD
F
![Page 59: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/59.jpg)
Can CarSpeak’s MAC assign fair share to content?
2 requested cubes, 10 moving robots
CD
F
0
0,2
0,4
0,6
0,8
1
0 1 2 3 4
Ratio of data transmitted from Cube 1 to Cube 2
802.11 – Req&Res
CarSpeak
![Page 60: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/60.jpg)
Can CarSpeak’s MAC assign fair share to content?
2 requested cubes, 10 moving robots
CD
F
0
0,2
0,4
0,6
0,8
1
0 1 2 3 4
Ratio of data transmitted from Cube 1 to Cube 2
802.11 – Req&Res
CarSpeak
CarSpeak’s content-centric MAC divides the medium fairly between requested content
![Page 61: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/61.jpg)
![Page 62: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/62.jpg)
![Page 63: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/63.jpg)
![Page 64: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/64.jpg)
Reaction to Objects in Blind Spots
Number of independent transmitters
0
20
40
60
80
100
1 2 3 4 5 6 7 8
Perc
enta
ge o
f ru
ns
rob
ot
sto
ps
on
tim
e
![Page 65: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/65.jpg)
Reaction to Objects in Blind Spots
Number of independent transmitters
0
20
40
60
80
100
1 2 3 4 5 6 7 8
Perc
enta
ge o
f ru
ns
rob
ot
sto
ps
on
tim
e
No Communication (percentage= 0)
![Page 66: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/66.jpg)
0
20
40
60
80
100
1 2 3 4 5 6 7 8
802.11–Req&Res.
Perc
enta
ge o
f ru
ns
rob
ot
sto
ps
on
tim
e
Robot unlikely to stop!
Reaction to Objects in Blind Spots
Number of independent transmitters
![Page 67: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/67.jpg)
0
20
40
60
80
100
1 2 3 4 5 6 7 8
CarSpeak
802.11–Req&Res.
Perc
enta
ge o
f ru
ns
rob
ot
sto
ps
on
tim
eReaction to Objects in Blind Spots
Number of independent transmitters
![Page 68: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/68.jpg)
Number of independent transmitters
0
20
40
60
80
100
1 2 3 4 5 6 7 8
CarSpeak
802.11–Req&Res.14x
Perc
enta
ge o
f ru
ns
rob
ot
sto
ps
on
tim
e
CarSpeak enables vehicles to better deal with hidden objects in blind spots
Reaction to Objects in Blind Spots
![Page 69: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/69.jpg)
Outdoor Testbed
Instrumented Yamaha car with laser sensors
Pedestrian crosswalk in campus-like environment
![Page 70: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/70.jpg)
Detecting a Pedestrian in Blind Spot
Pedestrians emerge from lobby
Lobby is a blind spot
Infrastructure sensors, some can send view of lobby
![Page 71: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/71.jpg)
-6
-4
-2
0
2
4
6
8
10
Distance from crosswalk when pedestrian appears (m)
1-2 m 2-4 m 4-6 m 6-8 m
Outdoor ResultsH
ow
far
fro
m c
ross
wal
kC
ar s
top
s (m
)
![Page 72: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/72.jpg)
-6
-4
-2
0
2
4
6
8
10
1-2 m 2-4 m 4-6 m 6-8 m
802.11
Outdoor Results
Distance from crosswalk when pedestrian appears (m)
Ho
w f
ar f
rom
cro
ss w
alk
Car
sto
ps
(m)
Car overshoots crosswalkIf it’s < 4m away whenpedestrian appeared
![Page 73: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/73.jpg)
-6
-4
-2
0
2
4
6
8
10
1-2 m 2-4 m 4-6 m 6-8 m
CarSpeak
802.11
Outdoor Results
Distance from crosswalk when pedestrian appears (m)
Ho
w f
ar f
rom
cro
ss w
alk
Car
sto
ps
(m)
![Page 74: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/74.jpg)
![Page 75: CarSpeak: A Content-Centric Network for Autonomous Drivingconferences.sigcomm.org/sigcomm/2012/slides/session6/04... · 2012. 10. 26. · Google’s Autonomous Car Robots for Disaster](https://reader035.vdocument.in/reader035/viewer/2022071106/5fe00345345c61532b550195/html5/thumbnails/75.jpg)
Conclusion
A communication system fully integrated with autonomous driving
Content-Centric approach to data access & MAC
Generally applies to collaborative robotics, virtual reality and virtual games