Realtime motion planning for Manipulator based on Configuration Space

1
Real-time motion planning for Manipulator based
on Configuration Space

• Chen Keming
• Cis Peking University

2
Main Contents

• Introduction
• My current work
• Future work and related work
• C-Space visualization for Teleoperation

3
Introduction

• Manipulator Motion Planning Problems
• Statement
• Compute a collision-free path for a manipulator
  among obstacles
• Inputs
• Geometry of manipulator and obstacles
• Kinematics of manipulator (degrees of freedom)
• Initial and goal manipulator configurations
  (placements)
• Outputs
• Continuous sequence of collision-free manipulator
  configurations connecting the initial and goal
  configurations

4
Introduction

• Tool Configuration Space

5
Introduction

• Framework

Manipulator representation
Configuration space formulation
Discretization
Graph searching
Obstacles representation
6
My Current Work

• Motivation Towards real-time Human-Robot
  Interaction in dynamic environment
• Application
• (Mobile based) Manipulator interacts with human
  without collision
• Dual-arm robot (Chen Fen,Ding Fu-qiangand Zhao
  Xi-fang Collision-free Path Planning of
  dual-arm Robot. ROBOT,vol.24,Mar.2002)

7
My Current Work

• Assumption
• The input data are readily available at any time
• Manipulator representation
• Cylinders
• Reduction to 3 joints
• Obstacles representation
• Cylinders
• Combination of main body and arms

8
My Current Work

• C-Space formulation
• Reduction to determine whether 2 cylinders
  collide in 3D W-Space

Case 1
Case 2
9
My Current Work

• Schematic

10
My Current Work

• Goal configurations formulation using inverse
  kinematics
• Discretization
• Joint 1 161, Joint 2 71, Joint 3 121

11
My Current Work

• Lazy C-Space computation due to
• Large numbers of points in C-Space(total
  1,383,151 points)
• Real-time process requirement
• Graph searching (A)
• Why use A
• Optimal and complete
• Objective values (expanding nodes, time)

12
My Current Work

• Speed up A
• OPEN is implemented as
• hash table
• priority list(implemented as Binary Heap)
• CLOSED is implemented as hash table

An example (collision checking points more than
30000)
List implementation
Hash table and Binary Heap implementation
13
My Current Work

• Result

14
My Current Work

• Dealing with dynamic environment
• A Replanner Plan by A using all the available
  information at the start.
• Start tracing the optimal path
• If there is a discrepancy between the initial map
  and the actual environment, update the new cost
  values for the corresponding arcs, run A again
  for planning between the current position and the
  goal.

15
My Current Work

• A Replanner shortcoming
• If the goal configuration is far away, little
  changes may force the planner to use A over the
  whole C-Space, although the changes in the
  optimal path may be small
• Hence, A replanner can be grossly inefficient
  computationally for real-time process

16
My Current Work

• Optimization --- Dynamic A(D) Stentz, 1994
• Functionally equivalent to A replanner
• Make local changes to the map and the resultant
  optimal path when a discrepancy between map and
  the environment is found
• Essentially prunes the graph search
• So, D could be a proper choice for optimization.
  But so far, it has only been used in mobile
  robotics to move a robot to given goal
  coordinates in unknown terrain Koenig, 2002.

17
D Algorithm
c(x1,x2)1
c(x1,x3)1.4
c(x1,x8)10000,if x8 is in obstacle,x1 is a
freecell
c(x1,x9)10000.4, if x9 is in obstacle, x1 is a
freecell
18
Goal
Gate
Start
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
Exam 1
35
Exam 2
36
My Current Work

• Compared with A replanner in our problem, D
  performance superior over A replanner

Checking points per replanning
37
Future work and related work

• Modify program, make it more robust with more
  experiments, speed up with more modifications.
• D Limitation
• D search from goal configuration, what if there
  are several goal configurations (its common in
  manipulator motion planning)?
• When the goal object is moving
• Current on-line planning methods using A based
  techniques focus on multi-directional search and
  parallel planning (Dominik HENRICH, Christian
  WURLL and Heinz WÖRN, 1998, etc )
• D should be adapted for our problems

38
Future work and related work

• Consult other D-like replanning algorithms (e.g
  D Lite Koenig, 2002 )
• Survey other real-time motion planning techniques
  in high dimensional C-Space
• Decomposition-based methods (Kavraki, 2001,
  Mediavilla, 2002, etc)
• Probabilistic roadmap based methods(most deal
  with static environment)

39
Future work and related work

• Use a more general 3D model to represent
  manipulator and obstacles
• Hierarchy structure
• Tree structure

40
Future work and related work

• Taxonomy

41
Future work and related work

• Experiment using real robot arm a challenging
  work

Images from cameras
Computer vision techniques
Model parameters
Motion planning
42
C-Space Visualization for Teleoperation

• Applications of C-Space Visualization
• Provide important qualitative information for
  mechanical design (E.Sacks, C.Pisula and
  L.Joskowicz Visualizing 3D Configuration Spaces
  for Mechanical Design. ).
• Evaluation of path planning methods
• Teleoperation (I.Ivanisevic and J.Lumelsky
  Configuration Space as a Means for Augmenting
  Human Performance in Teleoperation Tasks. IEEE
  Trans.Syst.Man,Cyber.,vol.30,pp.471-484,Jun.2000).
  

43
C-Space Visualization for Teleoperation

• Its easier for humans to handle motion planning
  problems in C-Space than in W-Space

44
C-Space Visualization for Teleoperation

• Challenges
• When the computer which generates C-Space data is
  not the same as the computer which receives
  humans input, C-Space data must be transfered
  through network
• C-Space data are too large
• 16171121 for my current implementation
• C-Space data change caused by dynamic
  environment, etc
• Poor network bandwidth

45
C-Space Visualization for Teleoperation

• So, C-Space data compression is necessary
• Additional work

Framework
C-Space Data
3D Models Data
3D Model Data Compression
C-Space for a Cylinder Object