Mrinal Kalakrishnan: Chomp Motion Planner
This document was uploaded by user and they confirmed that they have the permission to share it. If you are author or own the copyright of this book, please report to us by using this DMCA report form. Report DMCA
Overview
Download & View Mrinal Kalakrishnan: Chomp Motion Planner as PDF for free.
More details
- Words: 861
- Pages: 15
Smooth Motion Planning and Control Willow Garage Internship, Summer 2009
Mrinal Kalakrishnan University of Southern California
August 28, 2009
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
1 / 10
Overview
Smooth Motion Planning
Smooth Control
CHOMP - Covariant Hamiltonian Optimization for Motion Planning, Ratliff et. al., ICRA 2009.
Automatic spline trajectory generation from waypoints.
Implemented in the
spline_smoother ROS
chomp_motion_planner
package.
Usable by all motion planners. Implemented in the
ROS package.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
2 / 10
CHOMP Covariant Hamiltonian Optimization for Motion Planning, Ratliff et. al., ICRA 2009
Sampling based planners. . . Find feasible collision-free paths very fast. Typically produce jerky / redundant motion. Paths require post-processing, smoothing, optimization.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
3 / 10
CHOMP Covariant Hamiltonian Optimization for Motion Planning, Ratliff et. al., ICRA 2009
Sampling based planners. . . Find feasible collision-free paths very fast. Typically produce jerky / redundant motion. Paths require post-processing, smoothing, optimization.
CHOMP. . . Approaches the problem from a different perspective. Inherently generates and optimizes smooth trajectories. Uses gradient information to push trajectories out of collision
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
3 / 10
CHOMP: The Algorithm
Create a naive initial trajectory from start to goal. Define cost = trajectory smoothness cost + collision cost. Run gradient descent on this cost function. Not just regular gradient descent. . .
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
4 / 10
CHOMP: The Algorithm
Create a naive initial trajectory from start to goal. Define cost = trajectory smoothness cost + collision cost. Run gradient descent on this cost function. Not just regular gradient descent. . .
The key: Covariant gradient updates
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
4 / 10
CHOMP: The Algorithm
Create a naive initial trajectory from start to goal. Define cost = trajectory smoothness cost + collision cost. Run gradient descent on this cost function. Not just regular gradient descent. . .
The key: Covariant gradient updates Every update to the trajectory is ensured to be smooth. Obstacle costs obtained from signed distance field.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
4 / 10
CHOMP: Distance Fields
Euclidean Distance Transform in 3-D. Each cell contains distance to closest obstacle. Gradient information easily obtainable. Cartesian gradient converted into joint space using robot kinematics (Jacobian transpose).
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
5 / 10
CHOMP: Smooth Joint Limit Projection Typical L1 joint limit projections (clipping) destroy smoothness. Smooth the L1 projection using a covariant update.
Further technical details about CHOMP available in the paper: Nathan Ratliff, Matthew Zucker, J. Andrew (Drew) Bagnell, and Siddhartha Srinivasa. IEEE International Conference on Robotics and Automation (ICRA), May, 2009. Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
6 / 10
CHOMP: Demonstration (See Video) Table Bookshelf-top Manipulation
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
7 / 10
CHOMP: The Optimization Process (See Video)
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
8 / 10
Smooth Control using spline_smoother
A set of algorithms for converting a trajectory of waypoints into splines. Use case: Motion planners output a kinematic “path” of joint positions, expect this path to be executed smoothly. Currently contains the following three algorithms: Simple numerical differentiation Clamped cubic splines (constrained start/end velocities, continuous velocities and accelerations) Fritsch-Butland monotonic cubic splines
Optimization-based algorithms not yet available due to lack of BSD-licensed quadratic program solver. Currently works in conjuction with joint_waypoint_controller to provide a backwards-compatible joint trajectory interface, but with smooth execution.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
9 / 10
Smooth Control using spline_smoother
A set of algorithms for converting a trajectory of waypoints into splines. Use case: Motion planners output a kinematic “path” of joint positions, expect this path to be executed smoothly. Currently contains the following three algorithms: Simple numerical differentiation Clamped cubic splines (constrained start/end velocities, continuous velocities and accelerations) Fritsch-Butland monotonic cubic splines
Optimization-based algorithms not yet available due to lack of BSD-licensed quadratic program solver. Currently works in conjuction with joint_waypoint_controller to provide a backwards-compatible joint trajectory interface, but with smooth execution.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
9 / 10
Smooth Control using spline_smoother
A set of algorithms for converting a trajectory of waypoints into splines. Use case: Motion planners output a kinematic “path” of joint positions, expect this path to be executed smoothly. Currently contains the following three algorithms: Simple numerical differentiation Clamped cubic splines (constrained start/end velocities, continuous velocities and accelerations) Fritsch-Butland monotonic cubic splines
Optimization-based algorithms not yet available due to lack of BSD-licensed quadratic program solver. Currently works in conjuction with joint_waypoint_controller to provide a backwards-compatible joint trajectory interface, but with smooth execution.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
9 / 10
Thanks
Questions, comments?
I am reachable at: [email protected] and [email protected]
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
10 / 10
Mrinal Kalakrishnan University of Southern California
August 28, 2009
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
1 / 10
Overview
Smooth Motion Planning
Smooth Control
CHOMP - Covariant Hamiltonian Optimization for Motion Planning, Ratliff et. al., ICRA 2009.
Automatic spline trajectory generation from waypoints.
Implemented in the
spline_smoother ROS
chomp_motion_planner
package.
Usable by all motion planners. Implemented in the
ROS package.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
2 / 10
CHOMP Covariant Hamiltonian Optimization for Motion Planning, Ratliff et. al., ICRA 2009
Sampling based planners. . . Find feasible collision-free paths very fast. Typically produce jerky / redundant motion. Paths require post-processing, smoothing, optimization.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
3 / 10
CHOMP Covariant Hamiltonian Optimization for Motion Planning, Ratliff et. al., ICRA 2009
Sampling based planners. . . Find feasible collision-free paths very fast. Typically produce jerky / redundant motion. Paths require post-processing, smoothing, optimization.
CHOMP. . . Approaches the problem from a different perspective. Inherently generates and optimizes smooth trajectories. Uses gradient information to push trajectories out of collision
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
3 / 10
CHOMP: The Algorithm
Create a naive initial trajectory from start to goal. Define cost = trajectory smoothness cost + collision cost. Run gradient descent on this cost function. Not just regular gradient descent. . .
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
4 / 10
CHOMP: The Algorithm
Create a naive initial trajectory from start to goal. Define cost = trajectory smoothness cost + collision cost. Run gradient descent on this cost function. Not just regular gradient descent. . .
The key: Covariant gradient updates
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
4 / 10
CHOMP: The Algorithm
Create a naive initial trajectory from start to goal. Define cost = trajectory smoothness cost + collision cost. Run gradient descent on this cost function. Not just regular gradient descent. . .
The key: Covariant gradient updates Every update to the trajectory is ensured to be smooth. Obstacle costs obtained from signed distance field.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
4 / 10
CHOMP: Distance Fields
Euclidean Distance Transform in 3-D. Each cell contains distance to closest obstacle. Gradient information easily obtainable. Cartesian gradient converted into joint space using robot kinematics (Jacobian transpose).
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
5 / 10
CHOMP: Smooth Joint Limit Projection Typical L1 joint limit projections (clipping) destroy smoothness. Smooth the L1 projection using a covariant update.
Further technical details about CHOMP available in the paper: Nathan Ratliff, Matthew Zucker, J. Andrew (Drew) Bagnell, and Siddhartha Srinivasa. IEEE International Conference on Robotics and Automation (ICRA), May, 2009. Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
6 / 10
CHOMP: Demonstration (See Video) Table Bookshelf-top Manipulation
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
7 / 10
CHOMP: The Optimization Process (See Video)
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
8 / 10
Smooth Control using spline_smoother
A set of algorithms for converting a trajectory of waypoints into splines. Use case: Motion planners output a kinematic “path” of joint positions, expect this path to be executed smoothly. Currently contains the following three algorithms: Simple numerical differentiation Clamped cubic splines (constrained start/end velocities, continuous velocities and accelerations) Fritsch-Butland monotonic cubic splines
Optimization-based algorithms not yet available due to lack of BSD-licensed quadratic program solver. Currently works in conjuction with joint_waypoint_controller to provide a backwards-compatible joint trajectory interface, but with smooth execution.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
9 / 10
Smooth Control using spline_smoother
A set of algorithms for converting a trajectory of waypoints into splines. Use case: Motion planners output a kinematic “path” of joint positions, expect this path to be executed smoothly. Currently contains the following three algorithms: Simple numerical differentiation Clamped cubic splines (constrained start/end velocities, continuous velocities and accelerations) Fritsch-Butland monotonic cubic splines
Optimization-based algorithms not yet available due to lack of BSD-licensed quadratic program solver. Currently works in conjuction with joint_waypoint_controller to provide a backwards-compatible joint trajectory interface, but with smooth execution.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
9 / 10
Smooth Control using spline_smoother
A set of algorithms for converting a trajectory of waypoints into splines. Use case: Motion planners output a kinematic “path” of joint positions, expect this path to be executed smoothly. Currently contains the following three algorithms: Simple numerical differentiation Clamped cubic splines (constrained start/end velocities, continuous velocities and accelerations) Fritsch-Butland monotonic cubic splines
Optimization-based algorithms not yet available due to lack of BSD-licensed quadratic program solver. Currently works in conjuction with joint_waypoint_controller to provide a backwards-compatible joint trajectory interface, but with smooth execution.
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
9 / 10
Thanks
Questions, comments?
I am reachable at: [email protected] and [email protected]
Mrinal Kalakrishnan (USC)
Smooth Motion Planning and Control
August 28, 2009
10 / 10
Related Documents
Mrinal Kalakrishnan: Chomp Motion Planner
May 2020 6
Mrinal Dhar
June 2020 2
Planner
April 2020 23
Planner
May 2020 26
Planner
May 2020 28
Teacher Planner
May 2020 2More Documents from ""
Takayama Rss2009 Workshop
May 2020 6
Daniel Munoz: Indoor Object Detection
June 2020 1
Planar Objects And Articulation Models
May 2020 3
Mrinal Kalakrishnan: Chomp Motion Planner
May 2020 6
