pick_ik

pick_ik is an inverse kinematics (IK) solver compatible with MoveIt 2.

The solver is a reimplementation of bio_ik, which combines:

• A local optimizer which solves inverse kinematics via gradient descent
• A global optimizer based on evolutionary algorithms

Critically, pick_ik allows you to specify custom cost functions as discussed in this paper, so you can prioritize additional objectives than simply solving inverse kinematics at a specific frame. For example, you can minimize joint displacement from the initial guess, enforce that joints are close to a particular pose, or even pass custom cost functions to the plugin.

If you are familiar with bio_ik, the functionality in this package includes:

• Reimplementation of the memetic solver (equivalent to bio1 and bio2_memetic solvers)
• Reimplementation of the numeric gradient descent solvers (equivalent to gd, gd_r, and gd_c solvers)
• Fully configurable number of threads if using the global solver
• Cost functions on joint displacement, joint centering, and avoiding joint limits

For more details on the implementation, take a look at the paper or the full thesis.

Getting Started

To get started using pick_ik, refer to the following README files:

• Installation
• Usage
• MoveIt Tutorial

Changelog for package pick_ik

1.1.1 (2024-11-30)

• Initialize with the best seed instead of initial seed at population wipeout (#77)
• Fix timeout calculation if solution callback fails (#73)
• Remove incorrect override of tip_frames (#68)
• Contributors: Amal Nanavati, Sebastian Castro, Timon Engelke

1.1.0 (2023-12-13)

• Support continuous (unbounded) joints properly (#59)
• Run elite gradient descent in separate threads (#61)
• Contributors: Sebastian Castro

1.0.2 (2023-07-25)

• New options to control solution quality and performance

  • Option to rerun optimization until valid solution or timeout (#53)
  • Option to keep optimizing after a valid solution was found (#46)
  • Approximate solution cost threshold (#51) and joint jump threshold parameters (#42)
  • Position scale parameter (#47)

• Fix ordering of joint limits when loading robot model (#54)
• Fix use of solution callback (#48)
• Remove unnecessary preprocessor macro (#40)
• Contributors: Marc Bestmann, Sebastian Castro, Erik Holum

1.0.1 (2023-03-28)

• Set Werror through CMake presets (#39)
• Replace lower_bounds with gt_eq (#37)
• Upgrade with new pkgs to fix issue with ROS
• Target include subdirectory
• Update Catch2 version to 3.3.0
• Fix overriding of package

• Fix orientation calculation in cost function and frame tests (#31)

  • Fix orientation calculation
  • Update plugin return values
  • Remove redundant (and incorrect) joints bounds check
  • Use Eigen angular distance calculation

• Small grammar fixes (#28)
• Contributors: Sebastian Castro, Stephanie Eng, Tyler Weaver

1.0.0 (2022-12-08)

• pick_ik inverse kinematics plugin compatible with MoveIt 2
• Numeric gradient descent (local) solver
• Memetic algorithm (global solver), configurable for single or multi-threading
• Basic goal functions: joint centering, avoid joint limits, minimal joint displacement
• Support for position-only IK and approximate solutions
• Dynamic parameter switching at runtime using generate_parameter_library
• Docker devcontainer workflow for VSCode
• Contributors: Chris Thrasher, Sebastian Castro, Tyler Weaver