Interaction Forces and Internal Loads in Parallel Manipulators with Actuation Redundancy

1. Problem Statement

The paper addresses a critical gap in the control and analysis of redundantly actuated parallel manipulators. While the concepts of interaction forces (forces causing "squeezing" without motion) and internal loads (forces causing internal stress) are well-established for "grasp-like" systems (e.g., multifingered grippers, legged robots), their application to parallel manipulators has been flawed.

• The Core Issue: Existing literature often blindly applies methods developed for grasp matrices (where wrench equilibrium is defined by aggregating contact forces) to parallel manipulators.
• The Flaw: Parallel manipulators map joint torques to end-effector wrenches via Jacobian matrices, not a grasp matrix. This mapping involves a two-stage transmission: joint torques $\to$ transmission weights $\to$ applied wrenches.
• The Consequence: Previous studies claimed that the minimum-norm solution for joint torques (using the standard Moore-Penrose pseudo-inverse) generates wrench distributions free of interaction forces. The authors argue this is physically erroneous because minimizing the norm of the torque vector ($\tau$) is not equivalent to minimizing the norm of the applied wrench vector ($f$) due to non-orthogonal basis vectors and transmission weights. This leads to incorrect force distributions that still induce unwanted internal stresses.

2. Methodology

The authors propose a generalized framework that correctly adapts null-space wrench analysis to parallel manipulators by explicitly modeling the torque-to-wrench transmission.

A. Theoretical Foundation

• Distinction of Concepts:
  • Interaction Forces: Defined geometrically (Kumar & Waldron) as forces where the difference between any pair of forces is orthogonal to the line connecting their application points. This minimizes the Euclidean norm of the applied forces ($\min \|f\|^2$).
  • Internal Loads: Defined dynamically (Udwadia-Kalaba) based on the principle of least constraint. These are forces that generate a resultant wrench without inducing internal stress relative to a dynamically equivalent system of lumped masses. This minimizes a weighted norm ($\min h^T M^{-1} h$).
• The Transmission Model:
  The paper reformulates the wrench equilibrium equation for parallel manipulators:
  $$h_o = J^T K^{-T} \tau$$
  Where $J^T$ defines the geometry of constraint wrenches and $K^{-T}$ represents transmission weights.
  To analyze forces at the end-effector, the authors introduce a basis matrix $B$ such that the applied wrench vector $h$ is related to torques by:
  $$h = B K^{-T} \tau$$

B. The Generalized Solution (Metric Tensor Approach)

The authors demonstrate that minimizing the torque norm $\|\tau\|^2$ is incorrect for eliminating interaction forces. Instead, they derive a weighted Moore-Penrose pseudo-inverse that minimizes the actual wrench norm.

1. For Interaction Forces (Equilibrating Torques):
   The optimization problem is transformed to minimize the wrench norm subject to the torque constraint. This introduces a Riemannian Metric Tensor $W_e$:
   $$W_e = K^{-1} B^T B K^{-T}$$
   The solution for the equilibrating joint torque vector $\tau_e$ is:
   $$\tau_e = (J^T K^{-T})^\dagger_{W_e} h_o$$
   Where the weighted pseudo-inverse accounts for the non-orthogonality of the actuator wrench directions.

2. For Internal Loads (Manipulating Torques):
   Similarly, to eliminate internal loads, a weighted metric tensor $W_m$ is derived using the system's inertia matrix $M$:
   $$W_m = K^{-1} B^T M^{-1} B K^{-T}$$
   The solution for the manipulating joint torque vector $\tau_m$ is:
   $$\tau_m = (J^T K^{-T})^\dagger_{W_m} h_o$$

C. Static Determinacy Constraint

The paper notes a physical limitation: For a parallel manipulator to generate a valid set of forces free of interaction forces, each leg must be statically determinate (i.e., have enough actuators to control the force direction and magnitude independently). If a leg is under-actuated (e.g., a single-actuator RRR leg), it cannot generate forces in arbitrary directions, making the "interaction-free" solution mathematically impossible for certain wrenches.

3. Key Contributions

1. Identification of Literature Errors: The paper rigorously identifies that previous methods (e.g., scaling factor methods using standard pseudo-inverses) fail to eliminate interaction forces in parallel manipulators because they minimize torque norms rather than wrench norms.
2. Unified Framework: It provides a generalized mathematical framework that bridges the gap between grasp-like systems and parallel manipulators, preserving the distinct nature of Jacobian-based wrench mapping.
3. Explicit Synthesis Methods: The authors derive explicit formulas for synthesizing joint torque vectors that guarantee:
   • Zero Interaction Forces: By using the weighted pseudo-inverse with metric $W_e$.
   • Zero Internal Loads: By using the weighted pseudo-inverse with metric $W_m$.
4. Clarification of Terminology: The paper reinforces the distinction between "interaction forces" (geometric) and "internal loads" (dynamic), arguing they are often conflated in robotics literature.

4. Results (Case Study: 3-RRR Planar Manipulator)

The authors validated their method using a 3-RRR planar parallel manipulator with 6 actuators (2 per leg).

• Comparison: They compared the feasible force polygons generated by:
  1. The standard minimum-norm torque solution (existing literature method).
  2. The proposed weighted pseudo-inverse solution (new method).
• Verification:
  • For a specific end-effector wrench, the standard method produced a set of forces that violated the zero-interaction condition (forces did not align correctly to avoid squeezing).
  • The proposed method produced a set of forces that satisfied the geometric condition for zero interaction forces (Eq. 2).
• Internal Loads: In a second scenario with a non-isotropic end-effector, the authors showed that the equilibrating force distribution (zero interaction) and the manipulating wrench distribution (zero internal load) are distinct and require different torque vectors.

5. Significance

• Correcting Physical Misconceptions: The paper corrects a fundamental misunderstanding in the field: minimizing joint effort does not necessarily minimize internal stress or interaction forces in parallel robots.
• Improved Control and Design: The proposed methods allow for more accurate estimation of a manipulator's wrench capabilities. This is crucial for:
  • Singularity Avoidance: Properly utilizing redundancy to avoid singular configurations.
  • Load Balancing: Distributing loads evenly among actuators to prevent overheating or failure.
  • Precision Control: Ensuring that "squeezing" forces do not damage the object being manipulated or the robot structure itself.
• Future Applications: The explicit torque synthesis methods provide a foundation for real-time control algorithms and the design of next-generation redundant parallel robots.

In summary, this paper establishes that the wrench mapping in parallel manipulators requires a metric-weighted optimization approach rather than a simple Euclidean norm minimization of joint torques to correctly manage internal loads and interaction forces.