Input Decoupling of Lagrangian Systems via Coordinate Transformation: General Characterization and Its Application to Soft Robotics

Pietro Pustina, Cosimo Della Santina, Frédéric Boyer, Alessandro De Luca, Federico RendaIEEE Trans. Robotics 2024 · DOI

TL;DR

This paper introduces a novel coordinate transformation for Lagrangian systems that decouples actuator inputs, simplifying control design. This "collocated form" is shown to be achievable for a broad class of systems, including underactuated ones like soft robots, by identifying necessary and sufficient conditions for its existence.

What Problem Does This Paper Address?

Controlling complex mechanical systems, especially those with many degrees of freedom or underactuation (where there are fewer actuators than degrees of freedom), is a significant challenge. The "Input Decoupling" (ID) problem aims to find a way to control these systems such that each actuator directly influences only one specific motion or output. For Lagrangian systems, this often involves dealing with a configuration-dependent actuation matrix, making direct control difficult. Existing methods for input decoupling often require complex feedback transformations or are limited to specific system types. This paper seeks to find a general method to achieve input decoupling through a coordinate transformation, simplifying the control problem without needing complex feedback.

Methods

The core of the paper is the introduction of a "collocated form" for Lagrangian systems. This form is achieved through a coordinate transformation, $ \theta = h(q) $, where $ q $ are the original generalized coordinates and $ \theta $ are the new "actuation coordinates." The key idea is to find a transformation such that the system's dynamics in these new coordinates have an actuation matrix that is block diagonal, with an identity matrix in the top-left block.

The authors derive necessary and sufficient conditions for the existence of such a transformation, primarily based on the integrability of the passive output $ \dot{y} = A^T(q)\dot{q} $. If this passive output is integrable, meaning there exists a function $ g(q) $ such that $ \nabla g(q) = A^T(q) $, then the system is deemed "collocated," and the transformation $ \theta = g(q) $ (or a related function) can achieve the desired input decoupling.

They analyze this for three scenarios:

  1. Fully Actuated Systems ($ m=n $): A direct coordinate transformation $ \theta = g(q) $ is shown to achieve the collocated form $ \dot{\theta} = u $.
  2. Overactuated Systems ($ m>n $): The transformation decouples $ n $ inputs, resulting in a block diagonal actuation matrix of the form $ (I_n, A_{o,\theta}(\theta)) $.
  3. Underactuated Systems ($ m<n $): The transformation results in an actuation matrix of the form $ (I_m, 0_{(n-m) \times m}) $, meaning $ m $ inputs directly control $ m $ degrees of freedom, while the remaining $ n-m $ degrees of freedom remain unactuated but are still influenced by the system dynamics.

A crucial finding is that systems driven by thread-like actuators (like tendons or fluidic chambers) are always collocated, making them amenable to this control strategy.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and uses examples of mechanical systems (e.g., geostationary satellite, spring-actuated mechanism, tendon-driven joint, continuum soft robot) for demonstration.

Key Insights and Ideas

  • "Collocated Form" Simplifies Control: The core insight is that by transforming coordinates, we can achieve a system representation where actuators directly perform work on specific generalized coordinates. This "collocated form" dramatically simplifies control design by decentralizing the input influence.
  • Integrability is Key: The existence of this collocated form hinges on the integrability of the system's "passive output." This provides a clear, testable condition for whether a system can be brought into this simplified representation.
  • Thread-like Actuators are Naturally Collocated: A significant theoretical result is that systems actuated by thread-like mechanisms (tendons, fluidic chambers) inherently possess this collocated property. This explains why control strategies based on actuator length often work well for soft robots.
  • Underactuation is Handled Gracefully: The framework naturally extends to underactuated systems, where the transformation decouples the available inputs to control a subset of the system's degrees of freedom, leaving the others to evolve according to their internal dynamics.
  • Coordinate Transformation vs. Feedback: The paper emphasizes achieving decoupling through a coordinate transformation rather than complex state feedback, offering a more direct and potentially robust control approach.

Key Results

  • General Conditions for Input Decoupling: Necessary and sufficient conditions are derived for the existence of a coordinate transformation that achieves input decoupling for Lagrangian systems.
  • Characterization of "Collocated Systems": A new class of systems, termed "collocated," is defined based on the integrability of their passive output, guaranteeing the existence of the desired coordinate transformation.
  • Application to Soft Robotics: The theory is demonstrated to be directly applicable to continuum soft robots, which are often underactuated and driven by thread-like actuators.
  • Controller Extension: Existing control strategies for underactuated mechanical systems are shown to be extendable to the broader class of collocated systems.
  • Experimental Validation (Simulated): The paper includes simulations demonstrating that a controller designed in the "actuation coordinates" (using the P-satI-D regulator) achieves superior performance in shape regulation for a soft robot compared to a controller designed in the original configuration space. Specifically, the P-satI-D regulator achieved a steady-state Cartesian error of $ 5.8 \times 10^{-5} $ m, while the PD+ regulator in q-space had an error of $ 1.2 \times 10^{-2} $ m.

Limitations and Future Work

  • Integrability Assumption: The core of the method relies on the integrability of the passive output. While this is shown to hold for thread-like actuators, it might not be guaranteed for all systems. The paper notes that if integrability doesn't hold, input transformations might be necessary.
  • Unactuated Coordinates: For underactuated systems, the choice of unactuated coordinates is arbitrary, as long as the Jacobian remains nonsingular. Future work could explore specific choices of these coordinates to further simplify the system's equations of motion.
  • Experimental Validation: While simulations are presented, the paper suggests that future work will focus on experimental validation of the proposed method.
  • Relaxing Integrability Hypothesis: The authors plan to investigate relaxing the integrability hypothesis, potentially at the cost of transforming the input, to broaden the applicability of the method.
  • Flatness and Output Conditions: Future work will also explore conditions for input-state or input-output exact linearization and the existence of flat outputs for the system.

Summary updated

BibTeX
@article{pustina2024input,
  author = {Pietro Pustina and Cosimo Della Santina and Frédéric Boyer and Alessandro De Luca and Federico Renda},
  title = {{Input Decoupling of Lagrangian Systems via Coordinate Transformation: General Characterization and Its Application to Soft Robotics}},
  journal = {IEEE Trans. Robotics},
  year = {2024},
  doi = {10.1109/tro.2024.3370089},
}
Figure