PoseDec: Analytic Pose Decoupling in SLAM

PoseDec introduces a unified theoretical framework for analytic pose decoupling in Simultaneous Localization and Mapping (SLAM). It formulates environmental information geometric constraints that are entirely sensor-agnostic and universally applicable to any pose parameterization method.

By abstracting the geometric constraints from the sensor hardware and the specific mathematical representation of the pose, PoseDec offers a robust mathematical foundation that easily plugs into a wide variety of downstream SLAM and computer vision tasks. We can take our previous work DCReg as just an application for ICP in degenerate environment. In this work, we aim to generalize this analytical approach to any scenario where geometric constraints can be explicitly modeled.

Key Features

• Sensor-Agnostic: Operates independently of the specific sensor modality (e.g., LiDAR, Camera, Radar, IMU).
• Universal Parameterization: Fully compatible with any state/pose parameterization method (e.g., Quaternions, Euler angles, Lie algebra $SE(3)$).
• Analytic Rigor: Provides a strict, mathematically grounded theoretical framework for analyzing environmental geometric constraints.

Applications

The PoseDec framework can be seamlessly integrated into various front-end and back-end modules, including but not limited to:

• Odometry: Enhancing state estimation accuracy across sequential frames.
• Loop Closure Detection: Improving the robustness of place recognition.
• Point Cloud Registration: Accelerating and stabilizing point-to-point, point-to-plane, or distribution-based matching.
• Visual Odometry (VO) Initialization: Providing reliable initial states for monocular or stereo setups.
• Perspective-n-Point (PnP): Optimizing camera pose estimation from 3D-2D correspondences.
• Extrinsic Calibration: Enabling high-precision spatial synchronization between multi-sensor suites.
• Radar SLAM

Getting Started

Prerequisites

• Ubuntu 20.04 (Recommended)
• C++14 or higher
• Eigen

Build

mkdir -p workspace/src && cd workspace/src
git clone https://github.com/JokerJohn/PoseDec
cd ..
catkin_make # or catkin build / colcon build

Examples

Point-level Observability analysis

ICP Robustness vs different perturbation of initial pose