Hierarchical planning 20194524 Jinhyeok Jang A. Online - - PowerPoint PPT Presentation

▶
hierarchical planning
SMART_READER_LITE
LIVE PREVIEW

Hierarchical planning 20194524 Jinhyeok Jang A. Online - - PowerPoint PPT Presentation

Mar 16, 2023 340 likes •635 views

Hierarchical planning 20194524 Jinhyeok Jang A. Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, Eduard Vidal et el. ICRA2019 B. Hierarchical Path Planner using Workspace Decomposition and

slide-1
SLIDE 1

Hierarchical planning

20194524 Jinhyeok Jang

A. Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, Eduard Vidal et el. ICRA2019 B. Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs, Geoge Mesesan et el. IROS2018

slide-2
SLIDE 2

2

  • Hierarchical planner

Review

  • The Maverick planner: An efficient hierarchical planner for autonomous

vehicles in unstructured environments, IROS 17

Two staged planner0.1 sec RRT*, 20 sec

slide-3
SLIDE 3

3

  • Hierarchical planner

Review

  • Dynamic Channel: A Planning Framework for Crowd Navigation, ICRA 19
slide-4
SLIDE 4

4

  • Hierarchical planner

Background

  • Usually use two kind of planner
  • Global planner
  • Gives guidance
  • Local planner
  • Find the actual path

Global Local Start Goal

  • bstacle

Global path Local path

slide-5
SLIDE 5

5

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

slide-6
SLIDE 6

6

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, ICRA2019 Vidal et el.

  • Introduction

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

  • Underwater vehicle
  • Autonomous Underwater Vehicle – AUV
  • Complex dynamics
slide-7
SLIDE 7

7

The Maverick planner: An efficient hierarchical planner for autonomous vehicles in unstructured environments, IROS 17

  • Introduction
  • Kinodynamic planning is required.
  • Too slow for online planning.

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

RRT*, 20 sec

slide-8
SLIDE 8

8

The Maverick planner: An efficient hierarchical planner for autonomous vehicles in unstructured environments, IROS 17

  • Introduction
  • Kinodynamic planning is required.
  • Too slow for online planning.
  • How about multilayered planning?

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

Two staged planner0.1 sec RRT*, 20 sec

slide-9
SLIDE 9

9

Planning Feasible and Safe Paths Online for Autonomous Underwater Vehicles in Unknown Environments, IROS2016 Hernandez et el.

  • Two sampling planner
  • Online RRT* as global planner
  • Online geometric path planning[1]
  • Can handle localization error
  • Finite horizon
  • Gives path fast
  • Asymptotically optimal
  • Sometimes gives infeasible path

Online RRT*, https://www.youtube.com/watch?v=A5-8LTBxbHQ

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

Online RRT*

slide-10
SLIDE 10

10

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, ICRA2019 Vidal et el.

  • Two sampling planner
  • SST as local planner
  • Stable Sparse RRT(SST) as local planner

Asymptotically optimal sampling-based kinodynamic planning, Li et el. IJRR 2016

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

SST’s data structure SST RRT*

slide-11
SLIDE 11

11

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, ICRA2019 Vidal et el.

  • Two sampling planner
  • SST as local planner
  • RRT* + SST

Asymptotically optimal sampling-based kinodynamic planning, Li et el. IJRR 2016

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

Planning time, RRT vs SST Total nodes, RRT vs SST

slide-12
SLIDE 12

12

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, ICRA2019 Vidal et el.

  • Two sampling planner
  • Using two planner
  • Global RRT* + Local SST

Asymptotically optimal sampling-based kinodynamic planning, Li et el. IJRR 2016

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

slide-13
SLIDE 13

13

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, ICRA2019 Vidal et el.

  • Result
  • Single SST vs RRT*+SST

Time

SST RRT*+SST

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

slide-14
SLIDE 14

14

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles, ICRA2019 Vidal et el.

  • Result
  • Real world Test

Online Multilayered Motion Planning with Dynamic Constraints for Autonomous Underwater Vehicles

128m path Consider water current Uniform 0.4m/s Not Consider water current Noisy water current

slide-15
SLIDE 15

15

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

slide-16
SLIDE 16

16

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Introduction

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Real world robots
  • Have very high dimensionality
  • Unavailable for C-space approach
  • Humanoids…

27 DoF humanoid

slide-17
SLIDE 17

17

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Introduction

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Handling high dimensionality
  • Probabilistic techniques
  • Task space
  • End- effector’s position, orientation, …

Workspace planning

slide-18
SLIDE 18

18

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Introduction

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Extract workspace information
  • Cell decompositionETC.
  • Parallel method
  • Another way of improving path planning algorithm.
  • Running multiple planners at the same time
slide-19
SLIDE 19

19

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Multilayered planner

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Global planner
  • Cell decomposition
  • Find collision free path.

Initial position Goal

slide-20
SLIDE 20

20

  • Multilayered planner

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Local planner
  • Task space RRT
  • Each polytope’s planner runs

parallel.

  • Expand to adjacent polytope

based on adjacency graph.

Initial position Goal

slide-21
SLIDE 21

21

  • Multilayered planner

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Local planner
  • Task space RRT
  • Each polytope’s planner runs

parallel.

  • Expand to adjacent polytope

based on adjacency graph.

Initial position Goal Unavailable

slide-22
SLIDE 22

22

  • Multilayered planner

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Local planner
  • Task space RRT
  • Each polytope’s planner runs

parallel.

  • Expand to adjacent polytope

based on adjacency graph.

Initial position Goal Cheaper

slide-23
SLIDE 23

23

  • Multilayered planner

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Final path
  • Global path connects the local planner’s path in workspace.

Initial position Goal workspace RRT’s tree End-effector follows the path

slide-24
SLIDE 24

24

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Comparison

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • 8 DoF redundant robot

C-space path length end-effector path length

𝑚𝑟 𝑚𝑞

slide-25
SLIDE 25

25

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Comparison

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • 9 DoF Humanoid reaching to the box

C-space path length end-effector path length

𝑚𝑟 𝑚𝑞

slide-26
SLIDE 26

26

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Comparison

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • 100 DoF hyper robot

C-space path length end-effector path length

𝑚𝑟 𝑚𝑞

slide-27
SLIDE 27

27

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs , IROS2018 Geoge Mesesan et el.

  • Comparison

Hierarchical Path Planner using Workspace Decomposition and Parallel Task-Space RRTs

  • Result
  • Suggested model always success 100% even for extreme case.
  • Shows shortest path both and .
  • Even single tree, matches/outperforms bi-directional planner.

𝑚𝑟 𝑚𝑞

slide-28
SLIDE 28

28

Quiz

  • Q1.
  • In first paper, authors use Three stage of planning

[True/False]

  • Q2.
  • In second paper, authors use Voronoi diagram for

global planning [True/False]