[Paper Review] Inverse Kinematics with Dual-Quaternions, Exponential-Maps, and Joint Limits

Benjamin Kenwright|arXiv (Cornell University)|Nov 2, 2022

Human Motion and Animation18 citations

TL;DR

This paper presents a real-time inverse kinematics solver for articulated characters using dual-quaternions and exponential-maps within a projected Gauss-Seidel iterative framework, enabling efficient, stable, and visually pleasing solutions with joint limits. The method achieves interactive performance (over 100Hz) and robust convergence even under extreme poses or unreachable targets.

ABSTRACT

We present a novel approach for solving articulated inverse kinematic problems (e.g., character structures) by means of an iterative dual-quaternion and exponentialmapping approach. As dual-quaternions are a break from the norm and offer a straightforward and computationally efficient technique for representing kinematic transforms (i.e., position and translation). Dual-quaternions are capable of represent both translation and rotation in a unified state space variable with its own set of algebraic equations for concatenation and manipulation. Hence, an articulated structure can be represented by a set of dual-quaternion transforms, which we can manipulate using inverse kinematics (IK) to accomplish specific goals (e.g., moving end-effectors towards targets). We use the projected Gauss-Seidel iterative method to solve the IK problem with joint limits. Our approach is flexible and robust enough for use in interactive applications, such as games. We use numerical examples to demonstrate our approach, which performed successfully in all our test cases and produced pleasing visual results.

Motivation & Objective

• Address the challenge of solving highly nonlinear, multi-degree-of-freedom inverse kinematics problems in real-time for complex articulated figures such as human hands and bipeds.
• Overcome limitations of traditional methods like Jacobian-based solvers by integrating dual-quaternions for compact rotation-translation representation and exponential-maps for smooth angular interpolation.
• Ensure robustness and stability under joint limits and ill-posed configurations (e.g., unreachable targets) through iterative projection and temporal coherency.
• Enable practical deployment in time-critical applications such as real-time games and VR by minimizing memory usage and computational cost.
• Demonstrate the feasibility and effectiveness of the Gauss-Seidel method as a flexible, low-overhead alternative to Newton or Damped Least Squares in complex IK systems.

Proposed method

• Represent articulated kinematic chains using dual-quaternions to encode both rotation and translation in a single mathematical object, enabling efficient concatenation and manipulation.
• Employ exponential-quaternion maps to parameterize joint rotations, providing smooth and continuous angular representations suitable for gradient-based optimization.
• Formulate the inverse kinematics problem as a system of nonlinear equations solved iteratively using the projected Gauss-Seidel method to enforce joint limits.
• Apply iterative projection during updates to constrain joint angles within user-defined limits, preventing invalid or unstable configurations.
• Leverage temporal coherency between frames by using the previous frame’s solution as a warm start, significantly reducing convergence time.
• Integrate the Gauss-Seidel solver with a Jacobian matrix that maps joint velocity changes to end-effector position and orientation changes, enabling local linear approximation of the kinematic chain.

Experimental results

Research questions

• RQ1 Can the Gauss-Seidel iterative method be effectively adapted to solve real-time inverse kinematics problems with dual-quaternions and exponential-maps?
• RQ2 How does the integration of joint limits via iterative projection affect the stability and convergence of the IK solver in complex articulated structures?
• RQ3 To what extent does temporal coherency improve performance and convergence speed in interactive applications?
• RQ4 How does the proposed method compare in terms of robustness and visual quality when handling unreachable targets or highly constrained poses?
• RQ5 Can the method maintain real-time performance (e.g., >100Hz) across diverse and unpredictable motion sequences, such as chaotic finger movements?

Key findings

• The proposed method achieved real-time performance with consistent frame rates above 100Hz on a standard desktop system, demonstrating suitability for interactive applications.
• The solver converged stably even when end-effectors were placed out of reach, producing a best-effort reach pose without oscillation or jittering.
• For small, incremental movements, the solver required only 1–2 iterations; for larger or more complex changes, convergence took up to 10–15 iterations, with total computation time under a few milliseconds.
• The method successfully handled a wide range of complex, unpredictable poses—including random finger movements—demonstrating robustness and flexibility.
• When joints reached their angular limits, convergence slowed slightly, but the solver remained stable and did not diverge.
• The algorithm demonstrated visual plausibility and smooth motion, even in cases with high degrees of freedom and non-intuitive configurations, confirming its practical viability for animation and game engines.