IKFlow: Generating Diverse Inverse Kinematics Solutions

TL;DR

This paper presents IKFlow, a deep learning-based method that efficiently generates diverse inverse kinematics solutions for robotic arms, enabling quick and accurate sampling of multiple joint configurations.

Contribution

IKFlow introduces a neural network approach to produce diverse inverse kinematics solutions rapidly, improving over traditional methods that typically yield only a single solution.

Findings

Generates 2000 solutions in under 10ms

Achieves accuracy within 10mm and 2° of exact solutions

Allows rapid refinement with classical methods

Abstract

Inverse kinematics - finding joint poses that reach a given Cartesian-space end-effector pose - is a common operation in robotics, since goals and waypoints are typically defined in Cartesian space, but robots must be controlled in joint space. However, existing inverse kinematics solvers return a single solution pose, where systems with more than 6 degrees of freedom support infinitely many such solutions, which can be useful in the presence of constraints, pose preferences, or obstacles. We introduce a method that uses a deep neural network to learn to generate a diverse set of samples from the solution space of such kinematic chains. The resulting samples can be generated quickly (2000 solutions in under 10ms) and accurately (to within 10 millimeters and 2 degrees of an exact solution) and can be rapidly refined by classical methods if necessary.