20
Numerical Integration and Rigid Body Dynamics for Potential Field Planners David Johnson

Numerical Integration and Rigid Body Dynamics for Potential Field Planners

  • Upload
    edison

  • View
    35

  • Download
    0

Embed Size (px)

DESCRIPTION

Numerical Integration and Rigid Body Dynamics for Potential Field Planners. David Johnson. Recall. Define the world as a potential field Towards goal Away from obstacles. Robot Motion. Move down the potential field to minimum Think about contours Which way has zero change ? - PowerPoint PPT Presentation

Citation preview

Page 1: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Numerical Integration and Rigid Body Dynamics for Potential

Field Planners

David Johnson

Page 2: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Recall• Define the world as a potential field

– Towards goal– Away from obstacles

Page 3: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Robot Motion• Move down the potential field to minimum

– Think about contours• Which way has zero change?• Which way has maximum?• The directional derivative gives the rate of change

in a direction: Duf(x,y) = f(x,y) . ∇ u

– Move along the gradient

),,()(1 nq

UqUqU

Page 4: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Gradient Descent

• Evaluate the gradient, move along it some step size, repeat

)(1 nnnn qUqq

Page 5: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Difficulties

• Determining step size can be difficult

• Narrow canyons difficult

Page 6: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Gradient and Force• This step

is the same as

since force is the gradient of a potential field• This is just a non-Newtonian physics

simulation – replace velocity with acceleration

)(1 nnnn qUqq

tqFqq nnn )(1

Page 7: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Potential Field vs Physics Simulation

• Potential field planner can thus also be seen as a physics simulation

• Robot has state (its configuration and velocities)

• Goal and obstacle apply forces• Evolve the state of the robot

Page 8: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Physics of a Point• Based on ordinary differential equation (ODE):

• For physics examples, the f may be constant for x, as for gravity, might also be constant for t

• Initial value problem– Given start state, how does it change over time?– Path of a point over time is the integral curve

),( tf xx

Page 9: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Integral Curve

Page 10: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Common way to solve the ODE

• Euler’s Method

• Many problems– No correction, so errors build– May diverge!– Requires tiny time steps to be at all useful, so

is not efficient

ttttttftt

)()(),()(

xxxxxx

Page 11: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Where does error come from?• Look at Taylor series

• For taking time steps,

• So

nn

axnafaxafaxafafxf )(!)(...)(

!2)())(()()( 2

ttxta

0

0

nn

tntfttfttttftfttf )(!)(...)(

!2)())(()()( 020

00000

Page 12: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Euler error

• Euler uses only first two terms of Taylor series

• Error dominated by missing quadratic

Page 13: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Use higher order

• Runge Kutta integration– Evaluates at multiple places to cancel out

error• Implicit integration

– Has to invert small matrix, but allows big steps

Page 14: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Midpoint methoda) Compute a normal Euler

integration stepb) Move halfway along that

change and evaluate again for a half timestep

c) Move from the original spot in the direction given by b), but for a full timestep.

d) This is called a second order method

Page 15: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Determine Stepsize

• Fixed– Simulation is only as fast as worst case

• Measure change in taking a step vs 2 half steps

Page 16: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Stiff ODEs• Equations with large constants

– Example: keep x at zero

– Euler integration

– Diverges if

kxx

t

ttt

xtktkxxx)1(

1

kt

tk22

Page 17: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Stiff ODEs

• Can add energy into the system• Use friction to dampen the system

– Prevent out of control oscillations

Page 18: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Back to potential field planner• Robot is not a point

– Can we do this in configuration space?• Apply forces to multiple locations on robot

body (n particles on robot body)

– Force

– Torque

n

iiFF

1

n

iii Fr

1

Page 19: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Rigid Body Dynamics

• Integrate force to get new translational velocity, position

• Integrate angular acceleration to get angular velocity, orientation

• In 2D, just (x,y,q) and (x’,y’,q’)• 3D is more complicated

''q I

''mxF

Page 20: Numerical Integration and Rigid Body Dynamics for Potential Field Planners

Back to Our Robot

• Sample points on the robot– Compute distance to goal– Compute distance to obstacles– Compute forces from each

• Apply forces to each point– Accumulate total force, total torque

• Do Euler integration to find new position, orientation