a path planning algorithm to enable well-clear low...

22
A Path Planning Algorithm to Enable Well-Clear Low Altitude UAS Operation Beyond Visual Line of Sight Swee Balachandran National Institute of Aerospace, Hampton, VA Anthony Narkawicz, César Muñoz, María Consiglio NASA Langley Research Center, Hampton, VA

Upload: others

Post on 18-Oct-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

A Path Planning Algorithm to Enable Well-Clear Low

Altitude UAS Operation Beyond Visual Line of Sight

Swee Balachandran

National Institute of Aerospace, Hampton, VA

Anthony Narkawicz, César Muñoz, María Consiglio

NASA Langley Research Center, Hampton, VA

Page 2: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Outline

• Motivation

• Background

• Integrated system

• Results

• Conclusions

2

Page 3: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Motivation

• Off-the-shelf autopilot systems are highly capable, e.g.,

waypoint flight plan following, station keeping, geofence

containment.

• Emerging applications require Unmanned Aerial

Systems (UAS) to fly beyond visual line of sight

missions.

• Require technologies to maintain separation between

UAS while also enabling mission progress and satisfying

geofence constraints.

• Two complementary approaches: UTM (off-board) vs

onboard autonomy.

3

Page 4: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Related work - Separation assurance

• Several decades of research

interest in airborne

separation assurance.

• Original focus: prevent loss

of separation between

manned aircraft.

• Pilot plays pivotal role in the

timely execution of

maneuvers in accordance

with suggested resolutions.

• Examples: TCAS-II, ACAS-X,

DAIDALUS.

4

Image: Wikipedia: https://en.wikipedia.org/wiki/Traffic_collision_avoidance_system

Page 5: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Related work - Geofencing

• Related work has mainly focused on preventing an

autopilot or a remote pilot from violating fence

boundaries.

• Typically involve a simple return to home maneuver

without considerations on mission constraints.

• Examples:

– Safeguard: An Assured Safety Net Technology for UAS, Dill et al.

– Multi-Mode Guidance for an Independent Multi-Copter

Geofencing System, Stevens et al.

5

Page 6: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Related work – Path planning

• Flight planning:

– Optimal control formulation, e.g., Pontryagin’s minimum

principle, Dynamic programming.

– Geometric approach, e.g., Dubin’s path.

– Discrete search methods, e.g., A*, Dijkstra.

– Probabilistic search methods, e.g., PRM, RRT.

• Complexity increases with dynamic environments.

6

Page 7: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Objective

• Requirements for autonomous operation:

– Avoid other air traffic in the airspace.

– Satisfy geofences and obstacle constraints.

– Decision making capability to return to mission or initiate

replanning if necessary.

– Emphasis on formal verification.

– Computation speed.

• The primary contribution of this work is the integration of

several previously developed formally verified tools to

achieve the above functionality.

7

Page 8: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

DAIDALUS

• Detect and Avoid Alerting

Logic for Unmanned Aircraft.

• DAA reference

implementation established

by RTCA DO-365.

8

(Figure is notional)

Muñoz et al., DAIDALUS: Detect and Avoid Alerting Logic for Unmanned Systems, Proceedings of the 34th Digital Avionics Systems Conference (DASC 2015).

Page 9: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Detection Logic

Detection logic determines the time interval of loss

of well-clear (LoWC).

9

(Figure is notional)

Page 10: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Maneuver Guidance Logic

Maneuver guidance logic allows the pilot in control to maintain or recover

well-clear status:

• Separation assurance bands, i.e., ranges of maneuvers that lead to

intrusion in hazard volumes.

• Recovery bands, i.e., ranges of maneuvers that lead to well-clear

recovery without intruding a protected volume.

10

(a) Separation assurance bands (b) Recovery bands

Page 11: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

PolyCARP

• PolyCARP is a library containing functions for polygon related computations.

• Formally verified using Prototype Verification System (PVS).

• Uses ray casting to determine if a given point is inside/outside a geofence:– Outside, when even crossing.

– Inside, when odd crossing.

11

Ray casting

Narkawicz, A. and Hagen, G. E., “Algorithms for collision detection between a point and a moving polygon, with applications to aircraft weather avoidance,” 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016.

Page 12: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Path planning

• Rapidly Exploring

Random Trees (RRT).

• Build a tree of feasible

paths.

• Sample the search space

randomly.

• Grow tree towards the

sampled node.

• Discard branches that

lead to conflicts.

12

RRT exploration

Page 13: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Problem Description

13

Page 14: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Data structures

• Aircraft state information

– Aircraft position in ℝ3.

– Aircraft velocity in ℝ3.

• List of traffic state information

– Traffic position in ℝ3.

– Traffic velocity in ℝ3.

• Parent node.

• List of children nodes.

• Each node is a snapshot of what the environment looks

like if a branch was taken.

14

Page 15: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Problem setup

Problem dynamics:

• 𝑋𝑛+1 = 𝑓(𝑋𝑛, 𝑈𝑛).

• 𝑋 = [𝑜𝑝, 𝑜𝑣, 𝑡𝑝, 𝑡𝑣].

• 𝑈 = 𝑣𝑟𝑒𝑓.

15

Page 16: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Constraint satisfaction

• Kinematic bands used to

eliminate tree expansion in

directions that lead to

conflicts.

• The current node and the

projected node are

checked for traffic conflicts.

• Branches leading to conflict

are discarded.

1616

Page 17: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Early termination heuristic

• At each step, the

algorithm checks to

see if the direct path to

goal is free from traffic

and geofence

constraints.

• Avoids unnecessary

tree expansion.

171717

Page 18: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Decision making

• Limited assumption.

• Uncertainty in traffic state measurement.

18

Page 19: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Results

19

Encounter scenario

Page 20: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Computation time comparison

• Capability to dynamically construct flight plans to

maneuver around other traffic and geofence.

• Computation on embedded devices that can be used by

UAS.

20

Iterations used Nodes explored Time taken (s)

Encounter 1 5 5 1.5572

Encounter 2 7 6 1.5802

Encounter 3 16 13 2.6758

(a) Computation time on a beagle bone (1GHz ARM® Cortex-A8)

Iterations used Nodes explored Time taken (s)

Encounter 1 7 5 0.1324

Encounter 2 19 14 0.2822

Encounter 3 10 7 0.1671

(b) Computation time on a Jetson TK1 (2.32GHz ARM quad-core Cortex-A15)

https://beagleboard.org/blackhttp://www.nvidia.com/object/jetson-tk1-embedded-dev-kit.html

Page 21: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

ICAROUS

• Implementation available in Java/C++ on Githubunder NASA’s Open Source Agreement.

• Current version integrates with the ArduPilot flight stack.

• Provides ground station support for visualizing kinematic bands.

• ICAROUS is a high level decision making framework enabling autonomy.

21

Page 22: A Path Planning Algorithm to Enable Well-Clear Low ...atmseminar.org/seminarContent/seminar12/presentations/12th_ATM… · autopilot or a remote pilot from violating fence boundaries

Conclusions and future work

• Extended detect and avoid capability for low altitude

UAS to account for other traffic and geofences.

• A local planner to quickly navigate around other UAS

and geofences to continue with mission.

• Some parameters require tuning based on area of

operation, mission speed, traffic speed, etc.

• Explore different sampling strategy to further speed up

computation.

• Incorporate “hover and wait” maneuvers to let other

traffic pass by before proceeding.

• Coordinating resolution among multiple aircraft.

22