Transcript University of Pennsylvania

University of Pennsylvania
Asymmetric Bandwidth Channel (ABC)
Architecture
Insup Lee
University of Pennsylvania
July 25, 1998
7/15/98
University of Pennsylvania
Principal investigators
•
•
•
•
•
Ruzena Bajcsy (PI), CIS
David Farber, CIS/EE
Vijay Kumar, MEAM/CIS
Insup Lee, CIS
Jonathan Smith, CIS/EE
7/25/98
University of Pennsylvania
The ABC Model
A is low latency and B/W, duplex
B is high latency and B/W, simplex
C is high latency and B/W, simplex
A may be real-time
Broadcast
C
B
A
Clients
7/25/98
Server
University of Pennsylvania
Motivation
•
•
•
•
Common communication paradigm
Becoming widely available: CATV, DBS, ADSL
Cost advantage for multicast
Applications
–
–
–
–
7/25/98
World-Wide Web (R/W>10)
Multiple robots
Shared virtual reality environment (a la “Snowcrash”)
Mobile computers
University of Pennsylvania
Outline (New Award)
•
•
•
•
The ABC model
Network resource management
Programming paradigm and support
Applications (robotics, virtual environment)
7/25/98
University of Pennsylvania
ABC Network Management
• Real-time/interactivity
• Integration with existing protocols (IP)
– protocol boosters
•
•
•
•
•
Routing algorithms
Resource multiplexing
Multicast as basic communication primitive
Mobile computing
Probabilistic real-time guarantees for wireless
communication
7/25/98
University of Pennsylvania
The ABC Computation Model
• A client
– makes a request and the server may reply by broadcasting the
requested data
– filters out the broadcasted data
– needs to receive broadcasted data at predetermined time
• The server
–
–
–
–
–
–
7/25/98
determines whether to broadcast or not
clusters data and decide what to broadcast together
schedules the broadcast server, types of QoS attributes
adapts scheduling policy based on the system history
manages local storage at broadcast server
may be replicated for scalability
University of Pennsylvania
Multiple agent coordination
• A timed asynchronous system
– distributed agents need to coordinate, under timing
constraints, to perform the control task
– performance failures
– decisions should be consistent, valid and timely
• Approach
–
–
–
–
7/25/98
Timed synchronous communication
N-way timed synchronization
Timed atomic commitment
Majority timed atomic commitment
University of Pennsylvania
Run-time monitoring and checking
Formal verification
System
Spec
Requirement
Spec
Design
Checker
System
Implementation
Monitoring
Script
Event
Handler
Corrector
Implementation
System
Run-time Check
7/15/98
Filter
Communication
Checker/
Corrector
University of Pennsylvania
Fundamental Issues on Monitoring
• How does a monitor gather information from a running
system?
• How does the monitor relate to requirements?
• How do we integrate dynamic monitoring with static
analysis?
• Can monitor be used to steer a system?
• What mathematical guarantees do monitors provide?
7/25/98
University of Pennsylvania
Coordinated Control of Robots
• Multiple robots with wireless
communication
Communication
• Tradeoff between local computation
and communication
• Scalable number of robots
• Simulation and implementation
• Apply hybrid systems modeling
– Analyze the effectiveness of coordinated
control
– Determine the sensitivities of control
parameters
Number of
robots
7/25/98
Computing/sensing
University of Pennsylvania
Experimental testbed
•
Three heterogeneous robots
– Robot 1
• Nomad (omni directional)
mobile platform
• fork-lift
– Robot 2
• Labmate (nonholonomic)
mobile platform
• Actively controlled robot arm
– Robot 3
• Labmate (nonholonomic)
mobile platform
• Passive arm
•
Each robot is controlled by two IBM
compatibles
7/25/98
•
Communication via wireless local area
network
– Proxim wireless LAN
• 1.4 Mbs (80 byte packages, 20 Hz)
– IPX or TCP/IP protocol
– Peer to peer network
– Scalable to 10 robots, 8 Hz
University of Pennsylvania
Planning and Sensing
• Framework
– One or more leaders in any
formation
– Leaders generate plans and
broadcast plans to followers
– Decentralized controllers
• Planning
– Planner generates trajectories
consistent with geometric,
kinematics and dynamic
constraints
– Currently only one leader
7/25/98
• Sensing
– Different robots have different
sensors
– Information from distributed
sources is integrated centrally
– Global map is broadcast
University of Pennsylvania
Reconstruction Goals and Mothod
• Goals
– robot localization
• central control of agents
• central integration of data
• use of maps
– local reconstruction
• dense depth information
• obstacle avoidance
• surface integration
• Method
– self-calibration
• avoids prior calibration
• additional constraints
• specialized optimization
• accurate for all motions
– stereo
7/25/98
University of Pennsylvania
Overview of VENUS
• Virtual Environment
Network Using Satellite
• Satellite for high BW
broadcast
• Allows global, consistent
updates
• Cost of satellite amortized
over all users
• Land-links for uni- or
multicast
7/25/98
Network
Cloud
University of Pennsylvania
VENUS
• A virtual environment network using satellite
• An architecture to support large-scale, wide-area
virtual reality
• client/server based
– clients send lightweight update packets to server via low-BW
links
– server collates and broadcasts them using satellite high-BW
link
• Advantages
– broadcast allows all users to view the entire world
– scalable since incremental cost per user is small
– lower load on server since it only acts as coordinator
7/25/98