Principled Domain Extension for Analytic IK

@inproceedings{cohn2026principled, title={Principled Domain Extension for Analytic IK}, author={Cohn, Thomas and Shaw, Seiji and Roy, Nicholas and Tedrake, Russ}, booktitle={RSS Workshop on The Geometry of Motion}, year={2026} }

Abstract

Motion planning algorithms often inherently model configuration space (C-space) as a Euclidean space. But when a robot must satisfy end-effector constraints during its motion (e.g. carrying an object with two hands), the set of valid configurations collapses to a measure-zero submanifold in the ambient configuration space. Recent work has examined the use of analytic inverse kinematics (analytic IK), a closed form mapping from end-effector space to joint angles, to construct charts with a large domain. By planning in that parameterized space, the constraint is eliminated, and the path adheres to the manifold by construction. The mapping is made well-defined by including self-motion parameters as input to resolve the many-to-one problem of IK.

Recent works have applied analytic IK to trajectory optimization under end-effector constraints. Decision variables describe a trajectory in parameterized space, costs and constraints are imposed in C-space, and gradient-based optimization is enabled by differentiating through the IK function. A key challenge is the limited domain of an IK function: nonlinear trajectory optimizers do not guarantee workspace reachability at each iteration of the optimization. Bespoke IK solutions made for particular manipulators can be made to return approximate solutions outside of the reachable workspace, but popular automatic IK solvers return "no solution" for non-reachable targets. This is catastrophic for general nonlinear optimizers.

We extend the domain of an analytic IK function by leveraging least-squares solutions, yielding a closed-form Jacobian for end-effector poses outside the reachable workspace. This preserves the gradient flow outside of the reachable workspace, allowing the optimizer to restore feasibility. Inspired by the numerical IK literature, we present efficient approximations to this Jacobian. Empirically, we demonstrate that the overall framework begins to close the gap between bespoke and automatically generated analytic IK solutions.

Poster