⚡ electrical engineering

Forward Kinematics Accuracy Enhancement of a Force-Torque Decoupled Spherical Parallel Manipulator near Singular Configurations within a Tripodal Robot

This paper presents an enhanced mechanical design for a force-torque decoupled spherical parallel manipulator that integrates an internal serial structure with a fourth joint sensor to significantly improve forward kinematics accuracy near singular configurations, validated through analytical derivation, simulation, and real-time prototype testing within a tripodal robot.

Original authors: David Feller

Published 2026-07-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: David Feller

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Robot with Three Legs

Imagine a robot that looks like a tripod. It has a central body (torso) and three legs. Unlike a normal robot that walks on wheels or has a fixed base, this robot is designed to do two things at once: walk (locomotion) and hold/manipulate objects (manipulation).

The "brain" of its movement is a special joint in the middle of its body called a Spherical Parallel Manipulator (SPM). Think of this joint like a ball-and-socket joint in your hip, but instead of one bone moving, it's a complex machine made of three arms working together to rotate the robot's body in any direction.

The Problem: The "Wobbly" Spot

Like any complex machine, this robot has a weakness. There are specific positions where the math used to figure out where the robot is pointing becomes very shaky. The authors call these "singular configurations."

The Analogy: Imagine trying to balance a broom on your finger. When the broom is perfectly straight up, it's stable. But if you tilt it just a tiny bit, it becomes incredibly sensitive. A tiny wobble in your finger (a small measurement error) causes the broom to swing wildly (a huge error in the robot's position).

In the robot's case, when it moves into these "wobbly" positions (like when it sits down or twists its legs sideways), the three sensors it usually uses to know its position aren't enough. A tiny mistake in reading the sensors gets amplified, making the robot think it's pointing in a completely different direction than it actually is. This is dangerous because the robot might try to move based on wrong information and fall over or drop what it's holding.

The Solution: Adding a Fourth "Eyes"

The paper proposes a clever mechanical fix. The robot's designers realized there was empty space inside the center of the robot's body. They decided to put a fourth sensor in that empty space.

The Analogy: Think of the robot's three main arms as three people trying to guess the direction of a wind. If they all stand in a triangle and guess, they might get it wrong if the wind is tricky. But if you add a fourth person standing right in the middle of the triangle, looking straight up, they can tell you exactly how much the wind is twisting.

This fourth sensor is attached to a special internal chain of joints (a "serial structure") that runs through the center of the robot. Crucially, this internal chain measures the exact type of movement that causes the robot to become "wobbly."

How It Works: The "Smart" Math

Just adding a fourth sensor isn't enough; you have to use the data correctly. The authors created a new mathematical method to combine the data from the three outer sensors and the one inner sensor.

The Analogy: Imagine you are trying to guess the temperature of a room.

The Old Way (3 Sensors): You ask three people standing near the windows. If the wind is blowing, their answers might be all over the place, and you get a confused average.
The New Way (4 Sensors): You ask those three people, but you also ask a fourth person standing right next to the thermostat in the center.
The "Dynamic" Twist: The authors realized that sometimes the fourth person is too loud. If the room is calm, you should listen mostly to the three people near the windows. But if the room is chaotic (near the "wobbly" spot), you should listen mostly to the person by the thermostat.

They programmed the robot to dynamically adjust how much it trusts the fourth sensor. When the robot is in a stable position, it ignores the fourth sensor a bit. When the robot gets close to the "wobbly" spot, it leans heavily on the fourth sensor to keep its balance.

The Results

The researchers tested this in two ways:

Computer Simulation: They simulated the robot moving and added "noise" (fake errors) to the sensors. They found that with the new fourth sensor and the smart math, the robot could figure out its position much more accurately, especially when it was in those tricky, wobbly positions. The errors were significantly smaller.
Real Robot: They built a physical prototype and tested it. The real robot confirmed that the new method works in the real world, allowing the robot to maintain its balance and accuracy even when it twists into difficult shapes.

Summary

The paper is about a robot with three legs that uses a special ball-joint in its body. The authors found that this joint gets confused in certain positions. To fix this, they added a fourth sensor inside the robot's core and invented a smart math trick that knows when to trust this new sensor the most. This makes the robot much more accurate and stable, preventing it from getting lost or falling over when it moves into difficult positions.

Technical Summary: Forward Kinematics Accuracy Enhancement of a Force-Torque Decoupled Spherical Parallel Manipulator near Singular Configurations

Problem Statement
Parallel manipulators, particularly Spherical Parallel Manipulators (SPMs), are prone to singular configurations where the relationship between joint space and task space becomes ill-conditioned. Near these singularities, small measurement inaccuracies in joint coordinates are amplified, leading to significant errors in the reconstructed end-effector (EE) posture. This degradation compromises control stability and accuracy, which is critical for applications requiring high precision, such as the legged locomotion and manipulation tasks proposed for the tripodal robot described in this work. While singularities are typically avoided, practical scenarios (e.g., a robot adopting a sitting posture) may necessitate traversing regions near these configurations. The paper addresses the specific challenge of enhancing forward kinematics (FK) accuracy in these ill-conditioned states without altering the fundamental mechanical conditioning of the manipulator.

Methodology
The proposed solution involves a novel mechanical design and a corresponding redundant kinematic algorithm:

Mechanical Design (Force-Torque Decoupling):
The authors present an extended SPM design that integrates an internal serial structure (an RRR chain) within the unused central space of the manipulator. This creates a hybrid 3-CCC (outer shell) + RRR (inner chain) mechanism.
- Decoupling: The outer 3-CCC linkage transmits torques, while the inner RRR chain absorbs forces, effectively decoupling force and torque transmission.
- Redundant Sensing: This internal structure allows for the attachment of a fourth joint sensor ( $E_4$ ) on the base axis of the inner chain ( $\hat{c}_1$ ). Crucially, this measurement axis coincides with the axis of rotation that becomes sensitive to singular configurations (specifically, the axis where a Type 2 singularity causes a gain in an uncontrollable Degree of Freedom).
Kinematic Formulation:
- Jacobian Derivation: A redundant sensor Jacobian matrix ( $J_{red}$ ) is derived, mapping velocities from the four input joints (three base actuators and the inner support axis) to the tool platform. This matrix is rectangular ( $3 \times 4$ ), allowing for the use of the Moore-Penrose pseudo-inverse.
- Forward Kinematics (FK) Solution: The FK problem is treated as an overdetermined optimization problem due to the redundant sensor data.
  - Step 1: A classical loop-closure approach using the three primary sensors ( $E_1$ – $E_3$ ) provides an initial estimate of the EE orientation using the Levenberg-Marquardt (LM) algorithm.
  - Step 2: A refinement step minimizes the error between the measured redundant sensor value ( $\tilde{Z}$ ) and the value predicted by the inverse kinematics of the estimated orientation. This involves a weighted least-squares minimization where the weight of the redundant sensor is adjustable.
Dynamic Attraction Control:
To prevent the redundant sensor from degrading accuracy in well-conditioned postures (where it might conflict with the primary sensors), the authors propose a dynamic adjustment of the attraction coefficient ( $k_4$ ) for the fourth sensor. This coefficient is modulated based on the manipulator's conditioning index ( $1/\kappa$ ). The system applies a higher attraction to the redundant sensor only when the manipulator approaches ill-conditioned states, thereby salvaging accuracy near singularities while maintaining stability elsewhere.

Key Contributions

Novel Mechanical Topology: The integration of an internal RRR chain within a 3-CCC SPM to achieve force-torque decoupling and enable the placement of a fourth sensor without occupying space on the spherical linkages.
Singularity Robustness: A method to explicitly measure the angular position of the singular DoF, allowing for the reconstruction of EE posture in configurations where the classical three-sensor method fails or becomes highly inaccurate.
Adaptive Algorithm: A two-step FK solver with a dynamically regulated sensor attraction coefficient that balances the influence of the redundant sensor based on the current mechanical conditioning.
Validation: Comprehensive analytical derivation, simulation with distorted measurement signals, and validation on a built physical prototype of a tripodal legged robot.

Results

Simulation: The study utilized simulated motion profiles approaching Type 1 and Type 2 singularities. Results demonstrated that the classical 3-sensor method exhibited exponential growth in orientation error as the conditioning index ( $1/\kappa$ ) approached zero. In contrast, the proposed 4-sensor method with dynamic attraction ( $4S_{ffk,D}$ ) maintained significantly lower error levels.
Error Reduction: In the most ill-conditioned regions, the dynamic 4-sensor approach reduced the maximum orientation error by a factor of approximately 3.45 compared to the 3-sensor method.
Statistical Distribution: While the dynamic 4-sensor setup showed a slightly wider spread in joint angle errors generally, it significantly reduced outliers and improved the interquartile range (IQR) for the directly measured axis ( $Z$ ), indicating more consistent tracking of ideal angles near singularities.
Prototype Validation: The method was implemented on a real-world tripodal robot prototype. Optical tracking analysis confirmed the simulation results, verifying that the dynamic adjustment successfully mitigated errors near singular configurations without compromising performance in well-conditioned states.

Significance and Claims
The paper claims that this work enhances the reconstruction accuracy of the forward kinematics problem for general SPMs subject to measurement inaccuracies, specifically near singular configurations. The authors emphasize that this improvement is achieved without altering the actual mechanical conditioning of the manipulator; rather, it mitigates the amplification of measurement errors that typically occurs in ill-conditioned postures.

The significance lies in enabling robust control for legged robots that utilize SPMs in their hip joints, allowing them to traverse postures (such as sitting or specific gait transitions) that would otherwise induce uncontrollable behavior or severe accuracy loss due to singularities. The approach is presented as generalizable beyond the specific tripodal robot, extending the typical application range of SPMs to tasks requiring high reliability in singular or near-singular regions, including medical and surgical contexts. The work validates that integrating redundant sensing into the unused internal space of a force-torque decoupled design is a viable strategy for overcoming the limitations of classical parallel manipulator kinematics.