Emerging Technologies in Cyber Physical Systems:
Vehicular Cloud Computing
Ashwin Ashok (Postdoctoral Research Associate with Prof. Peter Steenkiste)
[email protected]
http://www.andrew.cmu.edu/user/ashwina/
1. THE MODERN VEHICLE
2. BRIDGING THE GAP: Cloud Computing for Vehicular Applications
Diagnostics
Uploading measurements,
remote diagnosis, ..
In-vehicle Infotainment (IVI)
Navigation systems
Music, video, …
Vehicle IT Infrastructure
Customer
Elastic resource
Current “apps” are very basic
and built in (no “iCar” market)
Ever-growing appetite
for new apps/services
Safety Applications
Enabled by rich
Vehicular safety
Internet connectivity
services will get
a huge lift
Driving Assistance
Parking, platoon
driving, emergency
braking, …
Eventually: self-driving
Closed system with limited
resources (mostly used for storage)
Car lifetime is very long, e.g.,15 years
(upgrading the car is problematic)
Driven by two year lifetime
of smartphones
Cognitive Assistants
Gesture and voice recognition,
fatigue/attention
and more to
come..
Use rich internet connectivity to extend vehicular resources with a pool of elastic cloud resources
Challenges:
Partitioning applications across very heterogeneous
Runtime mapping of application components
3. TOWARDS A SOLUTION : Offloading Computation from OBU to Cloud
Wide range of issues
Computing:
On-Board Unit (OBU) and cloud,
have different strengths and
weaknesses
Networking:
cellular, WiFi, DSRC have different
properties and unpredictable
availability
Data:
Access to sensor and information
diverse across platforms
Must be: adaptive .. configurable ..
evolvable
Proposed design: Service Oriented Offloading
Example application
Vision based Gesture Recognition
Camera feed
Filtering
Feature extract
Tracking
Virtual machine and code migrations avoided
Classification
(machine learning)
Heterogeneous : any-OS friendly design
Scalable : Cloud services and app code are disjoint
Lower maintenance cost than dedicated hardware
Recognition
User Interface
4. VEHICULAR CPS:
Research Questions for
Vehicular Cloud Computing
Where should apps run?
Can they migrate? How does the
Execute in Execute in
Cloud
Vehicle OBU placement change over time, users,
devices?
Software
implementations
Require database
access
Algorithms get
updated all the time
Low payloads for
offloading
Govern app latency,
fidelity, storage
How is access to I/O devices managed?
Access mechanism, who is in control
CAMERA
How is access
to information managed?
Access to sensors, user preferences
Who is in charge and coordination?
What is displayed on the screen?
What are the outputs?