The Cartesian Hand: In-Hand Manipulation with All-Linear Fingers

General Robotics Lab | Duke University | Discovery Robotics Research

---

1. Introduction: Rethinking Dexterity

Traditional robotic hands typically mimic human anatomy, utilizing revolute joints (hinges) to create complex arcs of motion. While biologically inspired, this approach introduces significant mathematical complexity in the form of non-linear kinematics. The Cartesian Hand proposes a paradigm shift: replacing rotational joints with all-linear actuators.

By utilizing prismatic joints, the hand operates directly in Cartesian space, effectively eliminating simplifying the need for complex inverse kinematics (IK) solvers during high-precision in-hand manipulation.

Research Objectives

• Develop a finger mechanism based entirely on linear motion.
• Implement stable in-hand manipulation of rigid objects.
• Reduce the computational overhead of coordinate transformations.
• Validate the precision of Cartesian-based grasping.

---

2. System Architecture & Design

The core innovation lies in the finger design. Instead of a shoulder-knuckle-tip sequence of rotations, each finger is a coordinated system of linear slides.

Mechanical Configuration

The fingers are designed to move along orthogonal axes. This allows the fingertip position $\mathbf{p}$ to be defined simply as:

$\mathbf{p} = \begin{bmatrix} x \\ y \\ z \end{bmatrix}$

Where $x, y, z$ are the direct displacements of the linear actuators.

Key Insight: In a revolute system, a small change in joint angle $\theta$ produces a non-linear change in tip position. In the Cartesian Hand, a change in actuator length $\Delta l$ results in a direct, linear translation $\Delta p$.

Hardware Comparison

---

3. Kinematic Modeling

The movement of the Cartesian Hand can be described using basic vector addition. If the base of the finger is at origin $O$, the tip position is the sum of the linear extensions.

The Math of Motion

The transformation from the actuator space to the fingertip position is represented by the identity matrix in the simplest case:

$\begin{bmatrix} x \\ y \\ z \end{bmatrix} = \mathbf{I} \begin{bmatrix} d_1 \\ d_2 \\ d_3 \end{bmatrix}$

For manipulation, the hand must maintain a contact force $\mathbf{F}$ while translating the object. The relationship between the actuator force $f_{act}$ and the contact force is:

f_act = J^T * F

Where J is the Jacobian matrix. Because the joints are linear, J becomes a constant or near-constant matrix, drastically reducing the CPU cycles required for real-time control.

def calculate_tip_position(actuators):
    # Simple linear mapping for Cartesian Hand
    # actuators = [d1, d2, d3]
    x = actuators[0]
    y = actuators[1]
    z = actuators[2]
    return (x, y, z)

---

4. Manipulation Workflow

The process of manipulating an object (e.g., rotating a cube) involves coordinated linear shifts across multiple fingers.

Execution Steps

1. Grasp: Fingers close linearly until a force threshold is met.
2. Translate: Actuators move in unison to shift the object.
3. Rotate: Fingers move in differential linear patterns to induce torque.

Figure 1: Conceptual representation of the orthogonal linear slides.

---

5. Nuance and Performance Analysis

While the Cartesian Hand simplifies the math, it introduces different engineering challenges:

• Workspace Volume: Unlike revolute joints that can "sweep" large areas, linear fingers are limited by the physical length of the rails.
• Singularities: The system avoids the common "gimbal lock" or kinematic singularities found in 6-DOF revolute arms.
• Precision: Because there is no angular amplification of error, the precision is limited only by the linear encoder resolution.

Summary of Findings

The research demonstrates that by moving the complexity from the software (IK solvers) to the hardware (linear rails), we can achieve highly predictable and stable in-hand manipulation. This is particularly useful for tasks requiring sub-millimeter precision in industrial assembly.