A Framework for Combining Optimization-Based and Analytic Inverse Kinematics

Analytic and optimization methods for solving inverse kinematics (IK) problems have been deeply studied throughout the history of robotics. The two strategies have complementary strengths and weaknesses, but developing a unified approach to take advantage of both methods has proved challenging. A key challenge faced by optimization approaches is the complicated nonlinear relationship between the joint angles and the end-effector pose. When this must be handled concurrently with additional nonconvex constraints like collision avoidance, optimization IK algorithms may suffer high failure rates. We present a new formulation for optimization IK that uses an analytic IK solution as a change of variables, and is fundamentally easier for optimizers to solve. We test our methodology on three popular solvers, representing three different paradigms for constrained nonlinear optimization. Extensive experimental comparisons demonstrate that our new formulation achieves higher success rates than the old formulation and baseline methods across various challenging IK problems, including collision avoidance, grasp selection, and humanoid stability.

💡 Research Summary

This paper introduces a unified framework that combines analytic inverse kinematics (IK) with optimization‑based IK by treating the analytic solution as a change‑of‑variables mapping. In traditional optimization‑based IK, the joint vector q is the decision variable and the constraint B X G(q)=T is a highly nonlinear equality that, when combined with additional nonconvex constraints such as collision avoidance or humanoid stability, often leads to poor convergence or outright failure. The authors propose to replace q with the end‑effector pose p (and optional self‑motion parameters s) as decision variables, using the analytic IK function q = f⁻¹(p, s) to recover joint angles. In this new formulation the original IK constraint becomes a simple linear equality (p = desired pose), while costs and constraints that were previously simple (joint limits, centering, collision avoidance) become more complex nonlinear expressions in the (p, s) space.

The paper details how this transformation aligns with the assumptions of three major families of constrained nonlinear solvers: interior‑point, augmented‑Lagrangian, and sequential quadratic programming (SQP). By moving the difficult equality constraint out of the interior of the feasible region, the solvers can operate on a problem that satisfies the Linear Independence Constraint Qualification (LICQ) and avoids the pathological behavior that arises when equality constraints are turned into pairs of inequalities. Automatic differentiation of the analytic IK mapping provides the required Jacobians for gradient‑based solvers.

Experimental evaluation covers three challenging IK scenarios: (1) collision‑free IK in cluttered environments, (2) grasp selection among multiple feasible grasps, and (3) static stability for a humanoid robot (Hubo). For each scenario the authors compare three solvers applied to the new formulation, the traditional “direct‑q” formulation, and a baseline Global‑IK method enhanced with modern free‑space segmentation. Across all tests the new formulation achieves markedly higher success rates—often 20–35 percentage points higher—especially on collision‑avoidance problems where the original formulation frequently fails. Runtime remains compatible with real‑time requirements, and the approach scales to robots with analytic solutions such as the PR2, KUKA iiwa, and Franka Emika Panda.

Additional contributions include refined analytic IK solutions for the PR2 bimanual manipulator and the Hubo humanoid, an improved objective function for collision‑free planning, and an inequality‑only representation of static stability that yields a feasible set of positive measure.

In summary, by leveraging an analytic IK solution as a smooth coordinate transformation, the authors convert a hard nonlinear equality into a linear one, thereby simplifying the landscape for generic nonlinear optimizers. The extensive empirical results demonstrate that this strategy consistently improves feasibility and robustness across diverse robots and tasks, suggesting a new standard for tackling complex, constrained IK problems in robotics.