Dynamically-stable Motion Planning for Humanoid Robots

1
Dynamically-stable Motion Planning for Humanoid
Robots

• Presenter
• Shen zhong
• Guan Feng
• 07/11/2003

2
Paper information

• Authors
• James Kuffner, Jr., Satoshi Kagami, Masayuki
  Inaba and Hirochika Inoue
• Address
• Dept. of Mechano-Informatics, The university of
  Tokyo
• http//www.jsk.t.u-tokyo.ac.jp/kuffner/humanoid

3
Outline

• Introduction of motion planning
• Motivation
• Robot model and problem
• Path search
• Statically-stable postures generation
• Experiments
• Discussions

4
Introduction

• Complete algorithms exist for general class of
  problem, but their computational complexity
  limits their use to low-dimensional configuration
  spaces
• Path planning methods using randomization are
  incomplete
• The goal is to develop randomized methods
• Converge quickly
• Simple enough to yield constant behavior
• Maintain robot static and dynamic stability

5
Motivation

• Develop a simulation environment to provide
  high-level software control for humanoid robot
• The software automatically computes object
  grasping and manipulation trajectories through a
  combination of inverse kinematics and randomized
  holonomic path planning

6
Motivation

• One part of the software is to design an
  algorithm for computing stable collision-free
  motions for humanoid robots given full-body
  posture goals

7
Difficulties

• High dimensions 30 or more

• Maintain overall static and dynamic stability

8
Solutions proposed

• Randomized planner
• RRT-Connect An efficient approach to
  single-query path planning. In proc.IEEE Intl
  Conf. on Robotics and Automation (ICRA2000), San
  Francisco
• Utilize Rapidly-exploring Random Trees (RRTs) and
  try to connect two search trees aggressively
• Filter the returned path to maintains dynamic
  balance constraints

9
Robot Model and Assumptions

• An approximate model of surrounding environment
  can be acquired using stereo vision or other
  means
• The robot is currently balanced on either one or
  both feet
• Supporting feet does not move during the planned
  motion
• A statically-stable full-body goal posture is
  given

10
Some notations

• Robot (A) with p links Li (i1,,p) is in
  workspace W. The ith link has mass ci relative to
  Cartesian frame Fi.
• A configuration of the robot is denoted by the
  set PT1,T2,,Tp
• n denotes the number of DOFs
• A configuration q is defined in C- configuration
  space
• The set of obstacles are labeled by B
• Cfree denotes the space of collision-free
  configurations
• X(q) denotes the vector representing the global
  position of the center of mass of A
• A configuration is statistically-stable if the
  projection of X(q) along the gravity vector lies
  within the area of support SP
• Cvalid denotes the subset of configurations that
  are both collision-free and statically-stable
• t I ? C denotes a motion trajectory,
  t(t0)qinitial, t(t1)qgoal

11
Path Search

• Path planner
• S.Kagami, F.Kanehiro, Y.Tamiya, M.Inaba and
  H.Inoue, Autobalancer an online dynamic balance
  compensation scheme for humanoid robots, March
  2000
• Planner(A,B,qinit,qgoal)? t
• Modified RRT-Connect try to connect two search
  trees aggressively

12
Path Search
e
qtarget
q
qnew
qnear
qinit
13
Path Search
14
Path Search
15
Statically-stable postures generation

• Many configurations are collision free but
  unstable.
• Many configurations q can be generated and stored
  in advance.
• Using collision detection algorithm.
• computing X(q) and verify that its projection
  along g is contained within the boundary of SP.

16
Statically-stable postures generation
17
Statically-stable postures generation
18
Statically-stable postures generation
19
Experiments
270 MHz SGI O2 (R12000) workstation DOF 30 or
more
20
Discussion and limitations

• The planner, having task-level planning
  algorithm, is limited to body posture goals and
  fixed position for either one or both feet.
• Reduction of computation time
• Efficient collision-detection software
• More stable samples
• Analysis of coverage of Cvalid and the
  convergence.

21

• Thank you !