This paper presents an online multi-query path planner for exploration tasks planned onboard an unmanned helicopter. While the desirable properties of roadmaps can be exploited in offline path planning, the dynamic nature of exploration scenarios hinders to utilize conventional roadmap planners. Hence, the presented path planning approach utilizes a deterministically sampled roadmap which is dynamically indexed in real time. To address situations of partial terrain knowledge, the roadmap can be extended from its a priori dimensions towards locations of unknown terrain that are outside its original, a priori boundaries. The multi-query property of the planning system allows for combinatorial optimization such that a rapidly acting decisional autonomy is achievable during exploration flights. D*-Lite is used as dynamic heuristic path searcher in order to re-plan efficiently. Inspired by the original work on this path search algorithm, the roadmap graph is augmented with an exploration vertex which steers the exploration behavior of the vehicle. As a result, the presented roadmap guides an unmanned rotorcraft through a priori unknown urban terrain in real time.
Multi-Query Path Planning for Exploration Tasks with an Unmanned Rotorcraft
1. Multi-Query Path Planning for Exploration Tasks
with an Unmanned Rotorcraft
Florian-M. Adolf
German Aerospace Center (DLR)
Institute of Flight Systems
Department of Unmanned Aircraft
Braunschweig, Germany
AIAA InfoTech@Aerospace 2012, Garden Grove, CA
Session: 26-I@A-19, Task Allocation and Planning Algorithms I
2. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Background
Support Acquisition of Situational Awareness in Hazardous Environments
www.DLR.de • Chart 2
Tepco Fukushima Daiichi Reactor, Japan 2011
[Air Photo Service + Rotomotion/Hélipse]
Earthquake, Chile 2010
Texas City disaster April 16, 1947:
Complex docks building.
[Special Collections, University of Houston Libraries]
3. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Problem Description
www.DLR.de • Chart 3
State of terrain
a priori unknown
Remote control link
may be disturbed when
flying out-of-sight
Intermediate paths
depend on acquired
terrain data
Repetitive path changes
UAV with terrain
mapping sensor
3-D structures with
overhangs might exist!
Autonomous Terrain Exploration
4. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Approach
www.DLR.de • Chart 4
Online Mapping and Multi-Query Path Planning
UAV with terrain
mapping sensor
“Raw” obstacle data
(e.g. point cloud, depth image)
Online Mapping
[Andert et al., 2009 / Krause 2010]
Geo-referenced
polygon obstacles
Online Path Replanning
[F.Adolf et al., 2010]
Path Following + Flight Control
[S.Lorenz et al., 2010]
Path updates
+ Replans efficiently on the way from „A to B“
+ Efficient multiple path queries
1. Roadmap expansion into unknown terrain
2. Decision making: „Best next“ waypoint „B“
Sensor
FOV
5. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Previous Roadmap-Based Path Planning
www.DLR.de • Chart 5
Roadmap
perimeter
B
A
Initial path
Polygon
updates
Non-traversable
roadmap edges
B
A
Replanned path
From “A to B”
What if we need to
extend the predefined
perimeter?
6. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Previous Roadmap-Based Path Planning
www.DLR.de • Chart 6
Initial roadmap
and its perimeter
1 B
Goal vertex
Acquired
during flight
B unreachable!
Non-traversable
roadmap edges
B
UAV
New obstacles
Issues Exploration from “A to B”
Resampling time hard to predict!
2b
Obstacle-based
resampling
B
UAV
7. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Roadmap Expansion Strategy
www.DLR.de • Chart 7
Initial roadmap
and its perimeter
B
Goal vertex
1
B
Resampled
‚unknown‘
partial volumes
A
UAV
New
obstacles
3
B
Increase chance
to find path:
Connection
strategy as for
initial roadmap
Exploration from “A to B”
8. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Autonomous Rotorcraft Testbeds
for Intelligent Systems (ARTIS)
www.DLR.de • Chart 8
Magnetometer
Power Supply
IMUGPSTelemetry
Flight Control
Camera
Computer Vision
Sonar
9. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Terrain Mapping
Geo-referenced
point cloud
Structure of interest
Area of interest
[Stefan Krause, 2010]
Remotely Piloted Aircraft System (RPAS)
www.DLR.de • Chart 9
10. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Simulation Setup
www.DLR.de • Chart 10
Closed Loop Flights in ‘Unknown’ Terrain
3-D LIDAR Model
50 m detection range
180 degree
scan plane 360 degree rotation @1Hz
of 2-D scan plane
Vehicle state update
ARTIS Closed Loop Simulation
Laser beam
collision detection
A Priori ‘Unknown’ Polygons
Extracted polygons
Path-based velocity
command
(VK, gamma, chi)
Roadmap-Based Planner
11. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
B
A
Exploration Scenario 1
www.DLR.de • Chart 11
Exploration from “A to B”
13. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Roadmap-Based Decision Making
www.DLR.de • Chart 13
Roadmap perimeter defines
volume to be mapped
A
Greedy Mapping: Select „Best Next“ Waypoint „B“
Uniform
edge costs
A
Bmap
Mapping
vertex
1 2
A2
Bmap
„Mapped“
vertices
A1
A0
Current „A to B“ path,
no path segment to Bmap
14. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Roadmap-Based Decision Making
www.DLR.de • Chart 14
Strategies to mark vertices as mapped:
1. Visited:
Physically passed or reached by the
vehicle.
2. Scanned:
All edges to and from a vertex have
been inside sensor FOV.
3. Uninformative:
If vertex is detected by mapping sensor,
it is not considered to provide useful
information anymore.
Greedy Mapping: Select „Best Next“ Waypoint „B“
proximity
radius
threshold
1) Visited
3) Uninformative
Mapping
sensor
FOV
All
edges
2) Scanned
Edge at least
once completely
within FOV
15. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
A
Exploration Scenario 2
Exploration of Urban Terrain
www.DLR.de • Chart 15
16. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Simulation Results
www.DLR.de • Chart 16
Rotating
LIDAR
sensor
UAV
Initial roadmap
perimeter
Exploration of Urban Terrain
17. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Simulation Result
www.DLR.de • Chart 17
Exploration of Urban Terrain
Efficient
replanning
Terrain almost
fully mapped
Total mission time within max. flight time of ARTIS
Trajectories
always well clear
of obstacles
18. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Summary
Two extensions to online roadmap-based path planner:
1. Roadmap Expansion for resolution completeness in ‘unknown’ volumes
while keeping number of samples low.
2. Online task planning roadmap:
1. Greedy mapping as example application
2. Replanning benefits from multi-query property
www.DLR.de • Chart 18
Ideas for improvements:
- Vertex marking strategy linked to real sensor instead of known FOV.
- Non-uniform edge costs:
Risk probability, account for GPS-denied locations, landing sites etc.
19. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Questions?
Thank you for your attention!
www.DLR.de • Chart 19
20. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Backup
www.DLR.de • Chart 20
21. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012
Simulation Result
www.DLR.de • Chart 21
Exploration of Urban Terrain
Remaining narrow corridor
(width < 20 m)
UAV
Rotating
LIDAR
Flown path
Current mapping path
22. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012www.DLR.de • Chart 22
Simulation Results
Influence of terrain detail level (1m)
23. > Multi-Query Path Planning for Exploration Tasks > Florian-M. Adolf • AIAA InfoTech, Session 26-I@A-19 > 20th June 2012www.DLR.de • Chart 23
Obstacle Detection and Mapping
[Andert et al., 2009]