Log-linear Backstepping control on $SE_2(3)$

Basic Information

• Paper ID: 2511.05775
• Title: Log-linear Backstepping control on $SE_2(3)$
• Authors: Li-Yu Lin (Purdue University), Benjamin Perseghetti (Rudis Laboratories), James Goppert (Purdue University)
• Classification: eess.SY (Systems and Control), cs.SY (Systems and Control)
• Submission Date: November 8, 2025
• Paper Link: https://arxiv.org/abs/2511.05775

Abstract

This paper proposes a log-linear backstepping control law on $SE_2(3)$ that preserves the complete rotation-translation coupling for rigid body systems evolving on nonlinear Lie groups. By leveraging hybrid invariant systems (group-affine dynamics models), the authors derive exact logarithmic error dynamics that are linear in the Lie algebra, without requiring local approximations. The paper provides closed-form expressions for the left and right Jacobian inverses of $SE_2(3)$, enabling exact error dynamics without local approximations. The log-linear backstepping control design ensures exponential stability of the error dynamics; due to the block-triangular structure of the error dynamics, linear matrix inequality (LMI) or $H_\infty$ gain performance design can be applied. This work establishes an exact backstepping control framework for hybrid invariant systems, providing a geometrically consistent theoretical foundation for future unmanned aerial vehicle (UAV) and spacecraft control design.

Research Background and Motivation

Core Problem

The motion of rigid body systems (such as UAVs and spacecraft) naturally evolves on nonlinear manifolds (such as $SO(3)$ or $SE(3)$), rather than Euclidean space. On these manifolds, traditional Euclidean subtraction loses geometric meaning, causing controllers designed in Euclidean space to exhibit inconsistent behavior.

Problem Significance

1. Geometric Consistency: Control of rigid body attitude and position must respect the underlying Lie group structure, ensuring control laws remain invariant under coordinate transformations
2. Coupled Dynamics: In practical systems, rotational and translational dynamics are coupled (e.g., gravity, thrust), requiring unified treatment
3. Global Validity: Avoiding singularity issues of local coordinate representations (e.g., Euler angles)

Limitations of Existing Methods

1. Local Approximations: Many methods use first or second-order approximations (e.g., $\log(\eta) \approx \eta - I$), with reduced accuracy for large errors
2. Decoupled Design: Treating attitude and position control separately, ignoring their inherent coupling
3. Lack of Exact Expressions: The Jacobian inverse of $SE_2(3)$ lacks closed-form expressions, limiting exact control design

Research Motivation

This paper aims to establish a geometrically exact, globally valid backstepping control framework, particularly for the extended Lie group $SE_2(3)$ containing velocity information, providing theoretically rigorous control design methods for systems like UAVs.

Core Contributions

1. Exact Logarithmic Error Dynamics: Derives exact logarithmic error dynamics for hybrid invariant systems on $SE_2(3)$, exhibiting linear structure in the Lie algebra without requiring local approximations
2. Closed-form Jacobian Inverse Expressions: Provides the first complete closed-form expressions for left and right Jacobian inverses of $SE_2(3)$ (9×9 block-triangular matrix form), including exact computation of translation kernels $Q_\ell$ and $Q_r$
3. Log-linear Backstepping Controller: Designs a backstepping control law preserving geometric structure, transforming the nonlinear system into block-triangular linear error dynamics
4. Exponential Stability Proof: Uses Lyapunov methods to prove exponential stability of the closed-loop system, providing explicit gain selection conditions
5. LMI/$H_\infty$ Compatible Framework: Due to the linear block-triangular structure of error dynamics, mature LMI and $H_\infty$ optimization tools can be directly applied for gain design

Detailed Methodology

Task Definition

Control Objective: Design control inputs (thrust $T$ and angular velocity $\omega$) such that system state $X = (R, v, p) \in SE_2(3)$ tracks reference trajectory $\bar{X} = (\bar{R}, \bar{v}, \bar{p})$, where:

• $R \in SO(3)$: attitude rotation matrix
• $v \in \mathbb{R}^3$: velocity in inertial frame
• $p \in \mathbb{R}^3$: position in inertial frame

Constraints: System dynamics follow hybrid invariant form (left-invariant + right-invariant terms)

Lie Group and Lie Algebra Foundations

$SE_2(3)$ Group Structure

Matrix representation: $X = \begin{bmatrix} R & v & p \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} \in SE_2(3)$

Corresponding Lie algebra $\mathfrak{se}_2(3)$: $[x]^\wedge = \begin{bmatrix} [\omega]_\times & a & v \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{bmatrix}$

where $x = [v, a, \omega]^T \in \mathbb{R}^9$

Invariant Error Definition

Left-invariant error: $\eta = \bar{X}^{-1}X \in SE_2(3)$

Explicit form: $\eta_p = \bar{R}^\top(p - \bar{p}), \quad \eta_v = \bar{R}^\top(v - \bar{v}), \quad \eta_R = \bar{R}^\top R$

Logarithmic error: $\xi = [\log(\eta)]^\vee \in \mathbb{R}^9$

Provides minimal coordinate representation of configuration deviation.

Hybrid Invariant System Dynamics

System dynamics model: $\dot{X} = (M - C)X + X(N + C)$

where:

• $M = [0, g, 0]^\wedge$: left-invariant term (gravity in inertial frame)
• $N = [0, Te_T, \omega]^\wedge$: right-invariant term (body thrust and angular velocity)
• $C$: kinematic coupling matrix (relationship $\dot{p} = v$)

$C = \begin{bmatrix} 0_{3\times3} & 0_{3\times3} & 0_{3\times1} \\ 0_{1\times3} & 0 & 1 \\ 0_{1\times3} & 0 & 0 \end{bmatrix}$

Exact Logarithmic Error Dynamics Derivation

Lemma 1 (Logarithmic Error Dynamics): Define $\tilde{M} = \bar{M} - M$, $\tilde{N} = \bar{N} - N$, then the logarithmic error satisfies:

$\dot{\xi} = -\text{ad}_{\bar{n}}\xi + ([\xi^\wedge, C])^\vee + J_r^{-1}(\xi)\tilde{n} + J_\ell^{-1}(\xi)\text{Ad}_{[\bar{X}^{-1}]}^\vee \tilde{m}$

where:

• $\text{ad}_{\bar{n}}$: adjoint action (generating cross-product terms from reference frame rotation)
• $J_\ell(\xi), J_r(\xi)$: left and right Jacobians
• $\bar{n} = [0, \bar{T}e_T, \bar{\omega}]^T$, $\tilde{n} = [0, \tilde{T}e_T, \tilde{\omega}]^T$

Explicit Form: $\dot{\xi} = \begin{bmatrix} -[\bar{\omega}]_\times & I & 0 \\ 0 & -[\bar{\omega}]_\times & -[\bar{T}e_T]_\times \\ 0 & 0 & -[\bar{\omega}]_\times \end{bmatrix}\xi + J_r^{-1}(\xi)\tilde{n} + J_\ell^{-1}(\xi)\text{Ad}_{[\bar{X}^{-1}]}^\vee \tilde{m}$

Closed-form Expressions for Jacobian Inverses

This is the paper's key technical contribution. First, expressions for the $SO(3)$ part:

$SO(3)$ Left Jacobian: $J_\ell^{SO(3)}(\omega) = I + \frac{1-\cos\theta}{\theta^2}W + \frac{\theta - \sin\theta}{\theta^3}W^2$

where $\theta = \|\omega\|$, $W = [\omega]_\times$

Left Jacobian Inverse: $S_\ell(\omega) = I - \frac{1}{2}W + \left(\frac{1}{\theta^2} - \frac{1+\cos\theta}{2\theta\sin\theta}\right)W^2$

Translation Kernels (key innovation): $Q_r(\omega) = \int_0^1 s R(s\omega) ds = q_0 I + q_1 W + q_2 W^2$

Coefficients: $q_0 = \frac{1}{2}, \quad q_1 = \frac{\sin\theta - \theta\cos\theta}{\theta^3}, \quad q_2 = \frac{1}{2\theta^2} - \frac{\sin\theta}{\theta^3} - \frac{\cos\theta - 1}{\theta^4}$

$Q_\ell(\omega) = J_\ell^{SO(3)}(\omega) - Q_r(\omega)$

$SE_2(3)$ Jacobian Inverse (9×9 block-triangular matrix): $J_\ell^{-1}(p,v,\omega) = \begin{bmatrix} S_\ell & -S_\ell Q_\ell S_\ell & -S_\ell Q_\ell(\omega; p)S_\ell \\ 0 & S_\ell & -S_\ell Q_\ell(\omega; v)S_\ell \\ 0 & 0 & S_\ell \end{bmatrix}$

where $Q_\ell(\omega; x) = (Q_\ell(\omega)x)_\times$ (tensor mapping)

Final error dynamics: $\dot{\xi} = (-\text{ad}_{\bar{n}} + A_C)\xi + \begin{bmatrix} c_3 & c_2e_T \\ c_4 & c_1e_T \\ c_1 & 0 \end{bmatrix}\begin{bmatrix} \tilde{\omega} \\ \tilde{T} \end{bmatrix} + \begin{bmatrix} d_2 \\ d_1 \\ 0 \end{bmatrix}\bar{R}\tilde{g}$

Backstepping Controller Design

Decompose error dynamics into three subsystems: $\dot{\xi}_p = -[\bar{\omega}]_\times\xi_p + \xi_v + c_3\tilde{\omega} + c_2e_T\tilde{T} + d_2\bar{R}\tilde{g}$$\dot{\xi}_v = -[\bar{\omega}]_\times\xi_v + [-\bar{T}e_T]_\times\xi_r + c_4\tilde{\omega} + c_1e_T\tilde{T} + d_1\bar{R}\tilde{g}$$\dot{\xi}_r = -[\bar{\omega}]_\times\xi_r + c_1\tilde{\omega}$

Step 1: Attitude Control Design virtual control $\xi_r^d$ (desired attitude error), angular velocity control law: $\tilde{\omega} = c_1^{-1}([\bar{\omega}]_\times\xi_r^d + \dot{\xi}_r^d - K_r(\xi_r - \xi_r^d))$

Define $e_r = \xi_r - \xi_r^d$, yielding: $\dot{e}_r = -[\bar{\omega}]_\times e_r - K_r e_r$

Step 2: Velocity Control Design virtual control $\xi_v^d$ and thrust $\tilde{T}$: $[-\bar{T}e_T]_\times\xi_r^d + c_1e_T\tilde{T} = -c_3\tilde{\omega} - d_1\bar{R}\tilde{g} - K_v e_v + \dot{\xi}_v^d - [\bar{\omega}]_\times\xi_v^d$

Define $e_v = \xi_v - \xi_v^d$, yielding: $\dot{e}_v = -[\bar{\omega}]_\times e_v - K_v e_v + [-\bar{T}e_T]_\times e_r$

Step 3: Position Control Design feedback linearization: $\xi_v^d = -c_3\tilde{\omega} - c_2e_T\tilde{T} - d_2\bar{R}\tilde{g} - K_p e_p$

Final closed-loop error dynamics: $\dot{\xi}_p = -[\bar{\omega}]_\times\xi_p - K_p\xi_p + e_v$$\dot{e}_v = -[\bar{\omega}]_\times e_v - K_v e_v + [-\bar{T}e_T]_\times e_r$$\dot{e}_r = -[\bar{\omega}]_\times e_r - K_r e_r$

Technical Innovation Points

1. Exactness: Completely avoids Taylor approximations of the logarithm map, valid throughout the Lie algebra
2. Block-triangular Structure: Backstepping design achieves cascaded structure of error dynamics, facilitating stability analysis
3. Geometric Consistency: All operations performed within Lie group/algebra framework, preserving coordinate invariance
4. Computability: All matrices ($c_1, c_2, c_3, c_4, d_1, d_2$) have explicit expressions

Stability Analysis

Theorem 4.1 (Exponential Stability)

Conditions: Gain matrices $K_p, K_v, K_r \in \mathbb{R}^{3\times3}$ are symmetric positive definite and satisfy: $\lambda_{\min}(K_r) > \frac{\|B\|^2}{2\lambda_{\min}(K_v)}$

where $B = -\bar{T}[e_T]_\times$

Conclusion: Equilibrium point $(\xi_p, e_v, e_r) = (0, 0, 0)$ is exponentially stable

Proof Strategy

Lyapunov Function: $V = \frac{1}{2}(\|\xi_p\|^2 + \|e_v\|^2 + \|e_r\|^2)$

Time Derivative: $\dot{V} = \xi_p^T\dot{\xi}_p + e_v^T\dot{e}_v + e_r^T\dot{e}_r$

Key Observation: Skew-symmetric terms $[\bar{\omega}]_\times$ satisfy $x^T[\bar{\omega}]_\times x = 0$, therefore: $\dot{V} = -\xi_p^T K_p \xi_p + \xi_p^T e_v - e_v^T K_v e_v + e_v^T B e_r - e_r^T K_r e_r$

Young's Inequality Bounds for Cross Terms: $\xi_p^T e_v \leq \frac{\kappa_p}{2}\|\xi_p\|^2 + \frac{1}{2\kappa_p}\|e_v\|^2$$e_v^T B e_r \leq \frac{\kappa_v}{2}\|e_v\|^2 + \frac{\|B\|^2}{2\kappa_v}\|e_r\|^2$

Final Bound: $\dot{V} \leq -\frac{\kappa_p}{2}\|\xi_p\|^2 - \frac{\kappa_v}{2}\|e_v\|^2 - \left(\kappa_r - \frac{\|B\|^2}{2\kappa_v}\right)\|e_r\|^2$

When gain conditions are satisfied, $\dot{V} \leq -2\alpha V$, thus: $V(t) \leq V(0)e^{-2\alpha t}$

proving exponential convergence.

Experimental Setup

Note: This is a theoretical control paper without numerical simulations or experimental validation. The paper focuses on:

1. Rigorous mathematical framework derivation
2. Complete closed-form expression provision
3. Theoretical stability proof

Potential Application Scenarios

The paper explicitly identifies the framework as applicable to:

• Unmanned Aerial Vehicle (UAV) Control: quadrotors, fixed-wing aircraft, etc.
• Spacecraft Control: attitude-orbit coupled control
• Other Rigid Body Systems: robotic manipulators, underwater vehicles, etc.

Implementation Considerations

Practical applications require:

1. Numerical computation of $Q_\ell, Q_r$ matrices (using series expansion when $\theta$ is small)
2. Gain optimization $K_p, K_v, K_r$ via LMI solvers
3. Real-time computation of reference trajectory Jacobian inverses

Related Work

Lie Group Control Methods

1. Geometric Control: Foundational work by Lee et al. on $SO(3)$ and $SE(3)$
2. Invariant System Theory: Left/right invariant error framework developed by Mahony, Jouffroy, and others
3. Backstepping Control: Successful applications of traditional backstepping in Euclidean space

Applications of $SE_2(3)$ Group

• Barrau & Bonnabel's Invariant Extended Kalman Filter (IEKF)
• State estimation in navigation and SLAM
• This paper's contribution: First systematic treatment of $SE_2(3)$ control problem

Jacobian Computation

• Barfoot & Furgale's Jacobian for $SE(3)$
• Sola et al.'s differential geometry toolbox
• This paper's contribution: First complete closed-form Jacobian inverse for $SE_2(3)$

Advantages of This Work

1. Exactness: Global valid formulas without approximations
2. Completeness: Complete chain from dynamics modeling to stability proof
3. Practicality: Compatible with mature tools like LMI/$H_\infty$

Conclusions and Discussion

Main Conclusions

1. Theoretical Completeness: Establishes complete backstepping control theory framework for hybrid invariant systems on $SE_2(3)$
2. Exact Expressions: Provides closed-form expressions for all necessary mathematical objects (Jacobian inverses, translation kernels)
3. Exponential Stability: Proves controller ensures exponential convergence of closed-loop system with explicit gain selection conditions
4. Engineering Usability: Block-triangular structure allows use of existing linear control tools (LMI, $H_\infty$)

Limitations

1. No Numerical Verification: Paper lacks simulations or experiments; actual controller performance unknown
2. Perfect Model Assumption: Assumes system perfectly follows hybrid invariant model; modeling errors not considered
3. Disturbance Robustness: External disturbance effects (wind, measurement noise) not analyzed
4. Computational Complexity: Real-time computation of 9×9 Jacobian inverse may impose computational burden
5. Singularity Handling: Requires series expansion when $\theta \to 0$; careful implementation needed

Future Directions

Implicit research directions indicated by the paper:

1. Numerical Simulations: Verify controller performance on quadrotor/spacecraft models
2. Robustness Extensions: Consider parameter uncertainty and external disturbances
3. Adaptive Control: Online estimation of unknown parameters (mass, inertia)
4. Obstacle Avoidance: Incorporate barrier Lyapunov functions for constraint handling
5. Optimal Control: Design $H_\infty$ or LQR gains using LMI framework

In-Depth Evaluation

Strengths

1. Mathematical Rigor

• Complete Lie group theory framework with rigorous derivations at each step
• Clear and complete proofs of lemmas and theorems
• Closed-form expressions eliminate approximation errors

2. Theoretical Innovation

• Originality: First complete expression of $SE_2(3)$ Jacobian inverse
• Unification: Handles attitude, velocity, position coupling within single framework
• Extensibility: Hybrid invariant system concept generalizable to other Lie groups

3. Methodological Elegance

• Backstepping design naturally exploits system's block-triangular structure
• Logarithmic error linearization in Lie algebra simplifies analysis
• Lyapunov proof direct and compact

4. Engineering Practicality

• Compatible with LMI tools for gain optimization
• Closed-form formulas facilitate real-time implementation
• Explicit stability conditions guide parameter selection

Weaknesses

1. Experimental Absence

• Major Deficiency: Complete lack of numerical simulations or experimental validation
• Cannot assess actual performance, convergence speed, robustness
• Computational complexity acceptability unclear

2. Strict Assumptions

• Requires system to precisely follow hybrid invariant form
• Real systems (e.g., quadrotors) have unmodeled aerodynamic drag, actuator dynamics
• Model mismatch effects not discussed

3. Insufficient Implementation Details

• No numerical stability handling for $\theta \to 0$ case
• Reference trajectory $\bar{X}(t)$ generation not discussed
• Practical gain selection guidance limited (only theoretical conditions)

4. Missing Comparative Analysis

• No comparison with existing methods (geometric control, sliding mode)
• Unclear how much practical advantage exact Jacobian provides over approximations

5. Writing Issues

• Symbol overloading ($c_1, c_2, c_3, c_4, d_1, d_2$ specific definitions require inference from equations 32-33)
• Lacks intuitive explanations and physical meaning
• No figures to aid understanding

Impact Assessment

Contribution to Field

• Theoretical Foundation: Provides rigorous mathematical basis for $SE_2(3)$ control
• Tool Value: Jacobian inverse formulas directly usable by other researchers
• Paradigm Shift: Promotes transition from approximate to exact geometric methods

Practical Value

• Medium-term Potential: Requires subsequent verification work before practical application
• Educational Value: Excellent teaching case for Lie group control
• Software Implementation: Could integrate into open-source projects like CogniPilot (related to authors' background)

Reproducibility

• Theory Reproducible: All derivations verifiable
• Implementation Reproducible: Lacks code and simulations; difficult to directly reproduce
• Needs Supplementation: Detailed algorithm pseudocode and numerical implementation guide

Applicable Scenarios

Highly Suitable

1. High-precision Spacecraft Control: Trajectory tracking with extreme accuracy requirements, ample computational resources
2. Theoretical Research: Starting point for further research (robustness, adaptivity, optimality)
3. Simulation Verification: Assessing control performance upper bounds under exact models

Suitable After Improvements

1. Real-time UAV Control: Requires computational simplification or efficient algorithms
2. Robust Applications: Needs extension to consider disturbances and uncertainty
3. Engineering Deployment: Requires complete implementation plan and debugging guide

Less Suitable

1. Low-cost Embedded Systems: Computational complexity potentially excessive
2. Highly Nonlinear Systems: Hybrid invariant assumptions not satisfied
3. Rapid Prototyping: Theory complexity may hinder quick development

Recommended Further Reading

Although the paper lacks a reference list, based on content, recommended readings include:

1. Lie Group Foundations:
   • Bullo & Lewis, "Geometric Control of Mechanical Systems"
   • Sola et al., "A micro Lie theory for state estimation in robotics"
2. Invariant Systems:
   • Mahony et al., "Nonlinear Complementary Filters on SO(3)"
   • Bonnabel, "Left-invariant Extended Kalman Filter"
3. Backstepping Control:
   • Krstic et al., "Nonlinear and Adaptive Control Design"
   • Lee et al., "Geometric tracking control of a quadrotor UAV on SE(3)"
4. $SE_2(3)$ Applications:
   • Barrau & Bonnabel, "Invariant Kalman Filtering"
   • Hartley et al., "Contact-Aided Invariant EKF"

Summary

This is a theoretically rigorous, mathematically elegant but experimentally unverified control theory paper. Primary value lies in:

✅ Theoretical Breakthrough: First complete mathematical framework for $SE_2(3)$ backstepping control
✅ Exact Expressions: Closed-form Jacobian inverse formulas represent important technical contribution
✅ Elegant Design: Logarithmic linearization + backstepping combination is natural
✅ Stability Guarantee: Rigorous proof of exponential convergence

❌ Verification Absent: Lack of any simulation or experiment is the major issue
❌ Implementation Details: Insufficient guidance for engineering implementation
❌ Comparative Analysis: Advantages over existing methods not quantified

Recommendation Index:

• Theoretical Researchers: ⭐⭐⭐⭐⭐ (Must-read; provides important mathematical tools)
• Engineering Practitioners: ⭐⭐⭐ (Await subsequent verification work)
• Student Learners: ⭐⭐⭐⭐ (Excellent Lie group control teaching case)

Future Expectations: Hope authors will soon release complete version with simulation verification, baseline comparisons, and actual flight tests to fully demonstrate practical value of this theoretical framework.