Autonomous Ground Systems Systems Architecture for Semantic - - PowerPoint PPT Presentation

Autonomous Ground Systems

8/9/2018

Systems Architecture for Semantic Information-Sharing in Unmanned Convoys

• Dr. Jeffrey L. Ferrin

Taylor C. Bybee Autonomous Solutions, Inc.

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 2

• Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2X) are common

abstractions of sharing information.

• Sensors like cameras and LiDAR have too much information to share over

network.

• Even if a network could handle that volume, the needed processing power
• n the receiving end would be astronomical!
• Solution?

– Achieve data compression by using mathematical models. – Rather than raw sensor data, share some semantic mathematic abstraction.

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

• Messaging occurs in broadcasts.
• Anyone in the network can send

and receive data.

• Information-sharing can be

designed as request-response messaging in pub-sub architecture.

8/9/2018 3

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 4

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

• Terrain can be represented in a

variety of ways.

• We choose to use a heightmap.
• We use tiled maps to represent

the areas.

• Heightmap cells have information

such as – Elevation estimate – Elevation estimate variance – Slope – Valid/Predicted/Invalid

8/9/2018 5

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 6

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 7

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys Tiles are used to represent heightmap and occupancy grid.

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 8

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Fusing heightmap and occupancy

grid tiles – Assume independent cells for a simple model – Use weights based on variance estimates, source of new information, timing of information, etc.

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 9

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Publish-Subscribe (ROS)

Implementation – Request-response communication needs to be in a separate thread so other callbacks remain uninterrupted. – Utilizes callback queues in each thread. – Mutexes are used to ensure thread-safety.

Thread Callback(s) Purpose 1 Main function Process point cloud data for terrain estimation. 2 Periodic Callback Broadcast a terrain request, wait, and process responses from the received queue. 3 Terrain Response Callback Receives responses, inserts these responses into the receiving queue. 4 Terrain Request Callback Receives request messages, publishes a response with any available data.

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 10

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Gazebo Simulator

– Two simulated Ford Escapes – Terrain with trees – Three trees of interest as

• bstacles

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 11

• With terrain-sharing, the terrain-obstacle segmentation algorithm should
• Have increased awareness of obstacles (from a farther distance)
• Have increased awareness of terrain (from a farther distance)
• Reduce false positives of obstacles

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 12

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Analysis

– Post-processed both terrain- sharing and no terrain-sharing scenarios. – Examined the number of allocated heightmap cells near each obstacle. – Examined the number of LiDAR points classified as non- terrain (obstacle) on each

• bstacle.
• Parameters

– Terrain is requested by each vehicle every 10 seconds.

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 13

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Obstacle #1 (First Tree)

– Significantly more LiDAR points are segmented as an

• bstacle farther away with the

terrain-sharing. – More heightmap cells are allocated at a farther distance with terrain-sharing.

Red – Without Terrain Sharing Green – With Terrain Sharing

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 14

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Obstacles #2 and #3

– Significantly more LiDAR points are segmented as an

• bstacle farther away with the

terrain-sharing. – More heightmap cells are allocated at a farther distance with terrain-sharing.

Red – Without Terrain Sharing Green – With Terrain Sharing

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 15

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Obstacle #1 (First Tree)

– Significantly more LiDAR points segmented as obstacle farther away with the terrain- sharing. – More heightmap cells are allocated at a farther distance with terrain-sharing.

Red – Without Terrain Sharing Green – With Terrain Sharing

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 16

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Vehicles Experiment

– Post-processed both terrain- sharing and no terrain-sharing scenarios. – Examined the number of allocated heightmap cells near each obstacle. – Examined the number of LiDAR points classified as non- terrain (obstacle) on each

• bstacle.
• Parameters

– Terrain is requested by each vehicle every 10 seconds. – Two Ford Escapes – VLP16 LiDAR – Ubiquiti Bullet Radios

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 17

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Ford Escapes
• Road and obstacle

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 18

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Obstacle #1 (First Tree)

– Approximately the same number of segmented LiDAR points for both cases. – Increased number of

• bserved heightmap cells

farther away. – Fewer false positives in adjacent area segmentation with terrain-sharing

Red – Without Terrain Sharing Green – With Terrain Sharing

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 19

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Obstacle #1 (First Tree)

– Approximately the same number of segmented LiDAR points for both cases. – Increased number of

• bserved heightmap cells

farther away. – Fewer false positives in adjacent area segmentation with terrain-sharing

No sharing Sharing

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.

Autonomous Ground Systems

8/9/2018 20

Systems Architecture for Semantic Information- Sharing in Unmanned Convoys

• Conclusions

– Preliminary results are promising – Need to be careful about how conflicting information is

• handled. (Race conditions.)

– Has been demonstrated at ASI for a manned explorer mapping obstacles with LiDAR, and passing to a follower with no sensors.

• Future Work

– Sharing other types of geo- spatial information. – Quantify how terrain-sharing can improve path control, lateral stability, and safety.

UNCLASSIFIED: Distribution Statement A. Approved for public release; distribution is unlimited.