Decentralized Intersection Management Shawn Adams Dr. Matthew - - PowerPoint PPT Presentation

Decentralized Intersection Management

Shawn Adams

• Dr. Matthew Rutherford

Department of Computer Science, University of Denver

Research Context

} Cooperative multi-agent path planning in and around

common geographic locations (e.g. an intersection)

} Using groups of autonomous vehicles navigating

intersecting waypoint paths

} Begin with structured intersections and paths

} e.g. two 2-lane roads that intersect at right-angles

} Interested in decentralized and distributed solutions } Initial research involves Peer to Peer directed diffusion } Expand into unstructured environments

} e.g. hazardous site investigation with geographic obstacles

Robot Main Loop - Centralized

start too close update pos & heading decelerate request reservation yes no arrived start

• ver

no inters- ection yes adjust heading & proceed no reser- vation yes yes stop moving no

Robot Main Loop - Decentralized

start too close update pos & heading decelerate attempt reservation yes no arrived start

• ver

no inters- ection yes adjust heading & proceed no reser- vation yes yes stop moving no process msgs

Decentralized Message Exchange

Robot Local State Peer List Reservations

3 7 2 4

Peer ID’s Peer ID’s

ID Tile Res Tile Res … ID Tile Res Tile Res … ID Tile Res Tile Res … … SN SN SN

Reservations and Conflicts

} A Reservation is

[ Robot ID, Sequence Number, [Tile, Tentry, Texit], … [Tile, Tentry, Texit] ]

} A conflict occurs when different robots need the same tile at the

same time and don’t know about each other

} Centralized: No conflicts allowed, single reservation list } Decentralized: Conflicts are detected locally while merging or

making new reservations

Robot 1 Tile 3 Tile 1 Tile 2

Time Tile 1

Tile 0 Robot 0 A Conflict bot 1 T0 T1 T2 T3 bot 0

Decentralized Conflict Resolution

} Two situations require conflict resolution

} Making a new reservation } Merging a peer’s list of reservations into my list of reservations

Make Reservation Merge Reservations

Form Reservation Get Conflicts Resolve Conflicts Add New Get Updated

Start End End Start

Decentralized - Conflict Resolution

Start End

Insert res res > all conflicts replace = true Cancel all conflicts yes replace = false no

‘res’ = the reservation to evaluate

} Conflicts detected using local

state

} Each robot assigned a priority

upon entering simulation

} Priority used in conflict

resolution

} All robots use the same

algorithm

} Given the same information, all

robots arrive at the same conclusion

Decentralized - Add New Reservations

Start End

conflicts = yours - mine iter ! = end() iter = conflicts iterator Get conflicts Resolve Conflicts no yes

Decentralized - Get Updated Reservations

Start

updates = yours U mine iter = updates iterator

End

iter ! = end() no y.seqNum > m.seqNum y = yours.find(iter) m = mine.find(iter) Get Conflicts conflicts. size == Update Reservation Resolve Conflicts no yes yes yes no

Simulation Environment

} Replicated Traffic Light and

Overpass results

} Good Built-in sensor suite } Provided ready to use platform to

test algorithms

} Simulation scalability limitations

Webots Commercial Simulator SSIM Discrete Event Simulator

} Implemented Overpass, Traffic Light,

Centralized Reservation, & Decentralized Reservation

} Very flexible } Provided scalability and speed we

needed

Simulation Results

Next / Current Steps

} Evaluate effect of realistic network conditions

} Network Latency } Dropped Packets

} Evaluate use of distributed hash table } Relax intersection constraints

} Unstructured Intersection } Robots from any direction

} Implement in actual robots

Questions?

Viewer Demonstration

Decentralized - Merge Peer Lists

Start

add = true

End

there is room add 1st new peer found no yes