Function-based Parametric Co-Design Optimization of Dexterous Hands

Mohammad Amin Mirzaee, Harsh Gupta, Wenzhen Yuan

University of Illinois at Urbana-Champaign

Abstract

Despite advances in dexterous hand manipulation, robotic hand design is still largely decoupled from task-driven evaluation and control, limiting systematic optimization. Existing robotic hand co-design approaches are often limited in scope, optimizing a small subset of design parameters. We introduce a comprehensive parametric framework for robotic hand generation that unifies palm structure, finger kinematics, fingertip geometry, and fine-scale surface curvatures within a single design space. Fine geometric features are introduced through parametric surface deformation kernels that directly influence contact interactions. We validate the framework on design optimization in grasp stability tasks in simulation and real-world dynamic scenarios. Our framework produces simulation- and fabrication-ready hand models and will be released as open-source to enable rapid design iteration for dexterous hand co-design optimization frameworks and cross-embodiment policy training and control research.

Framework Overview

The hand generation pipeline optimizes palm structure, finger configurations, surface geometry, and fingertip scale jointly. Each generated design is evaluated using physics-based grasp stability simulation, and results feed back into a Bayesian optimizer (TPE) to guide the next design iteration.

Sample

New hand designs are proposed by the TPE optimizer across 28 kinematic and geometric parameters.

Generate

A complete simulation- and fabrication-ready URDF is built for each candidate hand.

Evaluate

Grasp stability is scored via wrench-space tests on tool grasps simulated in Isaac Sim.

Iterate

Scores feed back to the optimizer, steering the search toward higher-performing designs.

Hand Generation

Palm Model

The palm geometry is parameterized as an extrusion of a planar 2D shape. Polygon sides, size, and aspect ratio define the initial outline. Finger and thumb bases are placed on the outline via continuous offset parameters, and the final mesh is extruded into assembly-ready palm bodies with cap and body components.

Finger & Thumb Model

Each finger is attached to its base on the palm and defined by a modular joint code. The code specifies rotation mode, optional side and axial extension joints, link lengths, and fingertip scale — enabling a large variety of kinematic structures from a single parameterization. The thumb uses a specialized configuration inspired by the Leap Hand, with an initial lateral joint for opposability.

Surface Geometry

Surface pads are generated on the base meshes and deformed using parametric Gaussian kernels that displace vertices along their surface normals. Multiple superimposed kernels produce complex but smooth contact surfaces — concavities, ridges, and bumps — that directly influence grasp contact forces and stability.

Simulation Results & SHAP Analysis

Co-design optimization over 28 parameters converges after ~200 iterations. SHAP analysis with a Random Forest surrogate reveals that finger number is the most influential parameter, followed by thumb normal offset and finger spread angle. Surface kernel parameters show nuanced opposing effects on grasp stability.

Optimization converges after ~200 iterations (A). SHAP analysis ranks finger number as the most influential parameter, followed by thumb normal offset and finger spread. Surface kernel parameters show nuanced opposing effects. Three hands of different scores — 0.81, 0.60, and 0.45 — are fabricated and evaluated (D).

Real-world Dynamic Experiments

Three fabricated hands are evaluated on a UR5e robot arm performing hammering, cutting, and mixing tasks. OptiTrack motion capture and robot end-effector poses are used to measure in-hand tool rotational slip. The highest-scoring simulation hand consistently outperforms alternatives across all three tasks.

Relative in-hand rotational slip is measured across cutting, hammering, and stirring tasks. The high-score hand consistently shows lower rotational error over time, validating the simulation-based optimization metric.

Citation

If you find this work useful, please cite:

@inproceedings{mirzaee2026handcdo, title = {Function-based Parametric Co-Design Optimization of Dexterous Hands}, author = {Mirzaee, Mohammad Amin and Gupta, Harsh and Yuan, Wenzhen}, year = {2026}, }