Networked Automotive CyberPhysical Systems

1
Networked AutomotiveCyber-Physical Systems
Prof. Rahul Mangharam Dept. Electrical Systems
Engineering University of Pennsylvania rahulm_at_seas
.upenn.edu
2
Outline

• Quick research overview
• Vehicle-to-Vehicle Wireless Networks
• Networked Automotive Active Safety
• Real-Time Congestion Probing

3
Todays Embedded Systems

• Deeply embedded in electronics
• Closed boxes
• Limited interaction with
• (Unpredictable) humans
• Environment
• Network of heterogeneous systems

Anti-Lock Brake System
4
Cyber-Physical Systems (1)

• Interaction with physical processes
• Closed-loop of Computation and communication
• Concurrent monitoring of multiple sensor sources
• Coordinated actions across a system of systems

5
Networked-CPS
Networks of Autonomous Vehicles
6
Networked-CPS

• Current networking techniques are inadequate
• Focused on moving data
• No concept of timing
• Embedded Networks
• Domain-specific protocols (CAN)
• Limited to centrally controlled nets

Such networking is the focus of this talk
7
Real-Time Embedded Wireless Networks
Large-scale Real-Time Programmable Systems for
Time-critical and Safety-critical applications

• A. Real-Time Sensor Network Protocols
• Predictable and Maximal Battery Lifetime
• Bounded End-to-End Latency
• B. Sensor Net Real-Time Operating System
• Explicit node network resource management
• Multi-platform and Scalable
• C. Platform Hardware
• Modular, multiple sensors, actuators, cameras
• Designed in-house with low-power operation

Global HW-based Time Synchronization
24-Ch EEG/ECG ASIC 25 mW
Multiple Sensors Audio, Light, Motion, Image,
Temp, Humidity
Smart Band-Aid 15 mW
8
Embedded Virtual Machines

• Real-Time Adaptive Middleware for Industrial
  Control Sensor Nets
• Distributed Runtime System for Parametric and
  Programmable control

9
Real-Time Embedded Wireless Networks
Active Networked Safety (Multi-hop V2V Safety
Alerts)
Closed-loop Wireless Fleet Coordination
Real-Time Congestion Prediction (V2V Networks
Algorithms)
Distributed RTOS and Real-Time Network Protocols
Modeling Middleware for Network Virtualization
Vehicle-to-Vehicle Networked Test-bed
10
Protocol Requirements for VANET
Bounded Broadcast
Scheduled Latency Storm
Flooding
Message Disconnected Adaptive
Persistence Network
Rebroadcast
End-to-End Rapid Topology Connectivity
Changes
Heterogeneous Networks
Alert Zone with delay-sensitive
messages
Warning Zone with persistent messages
11
GrooveNet Hybrid Network Simulator
12
GrooveNet - Hybrid Simulator
Vehicle Operations Director (VOD)
2
4
Cellular Link
Cellular Link
3
Real Vehicle
1
Real Vehicle DSRC Link
Virtual Vehicles (Simulated on VOD) V0 V1
V2 V3 V4 . Vn-1 Vn

• EVDO Uplink 200-300 vehicles/second
• EVDO Downlink 400-500 vehicles/second
• Multiple tests between Warren and Pittsburgh

13
(No Transcript)
14
GrooveNet Hybrid Simulator Design
Event Queue
GrooveNet Simulator Core
15
Modular Architecture
Traffic Light Model
Infrastructure Node Model
16
1,000 Vehicles in Chicago, IL suburb
Routed with Minimum Cost Routing
17
Minimum Weight Routing
Vehicles migrate to roads with higher speed limits
18
Performance Message Delay
19
Performance Message Lifetime
20
Experimental Multi-hop Vehicular Network Test-bed
5.9 GHz DSRC Dedicated Short Range Communications
Between vehicles
GPS
Differential GPS reference station beacons
Mobile Nodes in Pittsburgh, PA

• Vehicle-to-Vehicle Multi-hop
• Vehicle-to-Mobile Gateway
• Vehicle-to-Infrastructure

21
Driver Reaction Time

• Driver reaction time 1.5-2.5 seconds
• Slower reaction with higher cognitive load
• Driver Perception accounts for 50 of reaction
  distance
• Drivers respond faster to audio signals
• End-to-end Delay budget
• 1.5 sec for 1km
• 2.5 sec for 2km

22
GrooveSim On-Road Alerts (3)

• Only vehicles in the relevant geographic region
  receive alerts

23
Probabilistic Solutions to the Broadcast Storm
Problem

• Back-off Probability
• Location-based suppression
• Position-based suppression
• Neighbor-based suppression

• Limitations of Adaptive Broadcast Schemes
• Operating point selection is difficult
• Require relative and neighborhood information
• The trade-off between latency and link
  utilization is non-linear
• The bounds on the end-to-end latency are very
  loose
• Multiple flows cause Priority Inversion

How can we tighten the bounds on broadcast
latency?
24
Overview of LDMA Operation
Part I V2V LDMA-based Communication
Alert Zone

• Part I Fine-grained Synchronized Active
  Regions
• Region Definition shape with boundary
  coordinates
• Spatial Definition Block and Cell resolution
• Temporal Definition Slot schedule (µs fast time
  scale)
• Activity Lease Hours of operation and validity

25
Location Division Multiple Access
Assigned Black time slot
Assigned Red time slot
Embed Location-based slots in Map Database
26
Simple LDMA Schedule
Block 10 Block 11 Block12 ..
Block 0 Block 1 Block 2 .
Vehicle crosses Spatial block here
27
1-D LDMA Pipelines

• LDMA Spatial Definition
• 300m transmission range
• 100m LDMA cells
• LDMA Temporal Schedule
• A, C, B 200m/10ms ? 20,000m/s
• A, B, C 100m/10ms ? half-speed

28
Multiple LDMA Active Regions
Suburban Region A
Urban Region A
Downtown Region
Urban Region B
Suburban Region B
Rural Region A
29
Scalable LDMA Spatial and Temporal Representation

• Tree-based Cell activation and Schedule Assignment

30
LDMA 2D Scheduling

• County-wide slot assignment based on 2-D grid
  for dense urban regions
• Use 1D slot assignment for sparse rural roads and
  highways

31
LDMA Performance Comparison

• Trade-off between End-to-End Delay and Link
  Utilization
• 1D Chain of vehicles at 25 vehicles/km
• Adaptive Broadcast Schemes
• Neighbor-based best trade-off
• LDMA
• Smallest delay with controllable message receive
  rate

32
V2V Embedded Platform
GPS Active Antenna
Runs off 3AA batteries
802.11 radio
GPS receiver with PPS pulse for time
synchronization
GumStix 400MHz Linux computer
33
Time Synchronization Implementation

• Using GPS/PPS signal on a gumstix embedded
  computer
• Sub-200µs local synchronization accuracy with
  Linux 2.6
• 2ms pair-wise synchronization accuracy

34
LDMA Time Sync20ns GPS PPS Signal Jitter
35
LDMA On-Road Experiments
36
Real-Time Traffic Probing and Prediction
Algorithms
Historic Traffic Data
Datacenter
Large Number of Vehicles
37
Real-Time V2V Congestion Probing

• Centralized algorithm for fastest-path
  calculation
• Distributed algorithm for fastest-path
  calculation
• Dynamically update ETA
• Displays shortest and fastest path on map
• Displays congested routes on map

38
Different Paths
39
Different Paths
40
Different Paths
41
Different Paths
42
Different Paths
43
Different Paths
44
ETT Shortest Path Vs Fastest Path
45
Travel Time and Travel Distance
46
75 vehicles in Philadelphia
47
AUTOMATRIX

• 800,000 segments in greater D.C. area
• GrooveNet caps at 1,000 vehicles - takes
  overnight for 1hr test
• AUTOMATRIX - 5 million vehicles and more

48
Nvidia CUDA Programming Model
49
(No Transcript)
50
Traffic Incident Modeling
51
(No Transcript)
52
(No Transcript)
53
(No Transcript)
54
(No Transcript)
55
(No Transcript)
56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
Thank You
60
Why do we need GrooveNet?
61
Uniform Urban Distribution
62
Rural Area Rnd Waypoint Vs GrooveNetTopology-Mob
ility Models
63
Message Propagation Rate
64
GeoRoute Broadcast Scenarios
Highway Driving City
Driving Rural Driving

• Path with Intermediate points
• Static Source Routing

• Radial Broadcast

• Bounding Box
• Controlled Flooding

65
GrooveSim On-Road Alerts (1)

• Broadcast Safety Alerts to all vehicles in the
  vicinity
• Messages are valid in a specific geographic
  region
• Regions are determined by position, speed and
  direction

66
GrooveSim On-Road Alerts (2)
67
Active Region ProgrammingOut-of-band LDMA
Control Channel
Pilot Tone
Monaural Signal L R
FM Radio Band with RDS
Stereo Signal L - R
Sub-carrier Channel
Sub-carrier Channel
15 17 23 53
57 67
92 kHz

• Use FM/RDS (Radio Data System) with Open Data
  Channel
• A priori scheduling based on historical trends
• Reactive programming with on-road feedback
• At slower time scale (10s sec)
• Regional updates with Active Region Definitions