Rico贾若童的博客

Robotics - Gauss Newton (GN) and Levenberg Marquardt (LM) Optimizers

Newton's Method, GN, LM optimizers

Newton’s Method Solving An Equation To find an arbitrary equation’s root $f(x) = 0$, We start from an arbitrary point $x_0$ that’s hopefully close to the solution, $x_s$ The main idea of Ne...

Posted by Rico's Nerd Cluster on April 14, 2025

Robotics - [3D SLAM - 5] LIO Hands-On

Spatial Tiling

Issues That I Encountered IMU / LiDAR Synchronization There are several real-world timing problems to handle: The ROS publisher, bag writer, and LiDAR driver are all bursty — LiDAR data can a...

Posted by Rico's Nerd Cluster on April 8, 2025

Robotics - [3D SLAM - 5] Map Metrics

Gaussian Entropy, Graph Tension, Chi Squared, Trajectory Smoothness

Residual Calculation In scan matching, a residual is the raw error term e. The weighted squared error is: 1 chi2 = e.transpose() * information * e; where: 1 information = covariance.inverse() ...

Posted by Rico's Nerd Cluster on April 8, 2025

Underwater LiDAR: How Light-Based Sensing Works Beneath the Waves

Introduction LiDAR (Light Detection and Ranging) is incredibly effective in air, enabling autonomous vehicles, drones, and robots to map their surroundings with centimeter-level precision. But wha...

Posted by Rico's Nerd Cluster on April 7, 2025

Robotics - [3D SLAM - 6] KISS ICP Review

KISS-ICP: Simple Odometry, Done Carefully The paper’s core claim is simple: you do not need a complicated frontend, feature extraction, normals, surfels, IMU fusion, or learned components to get s...

Posted by Rico's Nerd Cluster on March 17, 2025

Robotics - [3D SLAM - 4] GICP

Generalized ICP, BFGS Optimization

1. Intuition In GICP (Segal et al, 2009), each matched pair is not just two points $p_s$ and $p_t$. Each point also carries a local covariance estimated from its neighbors (for example, 20 nearest...

Posted by Rico's Nerd Cluster on March 7, 2025

Robotics - [3D SLAM - 3] Lidar Odometry

Direct Lidar Odometry, NDT, Incremental NDT, Indirect Lidar Odometry, LOAM

Direct Lidar Odometry When given two scans, a direct NDT method means no features are extracted. The two scans would directly go through NDT for finding their relative poses. The workflow is as fo...

Posted by Rico's Nerd Cluster on March 4, 2025

Robotics - [3D SLAM - 2] Scan Matching Hands-On Notes

Preact-Mojave, Lidar-Only Front-End

Preact-Mojave Build & Workflow The ROS2 driver for Preact-Mojave can be found here. When testing it, I ran into below questions: Why make provision & make ros2? make provisi...

Posted by Rico's Nerd Cluster on March 3, 2025

Robotics - [3D SLAM - 1] Scan Matching

3D Point-Point ICP, 3D Point-Plane ICP, NDT

The Scan Matching Problem When we run scan matching, we usually have two point clouds: a source scan $S_1 = {\mathbf{p}_1, \ldots, \mathbf{p}_n}$ (the new measurement), and a target scan $S_...

Posted by Rico's Nerd Cluster on March 1, 2025

Robotics - [Bugs - 2] SLAM Related Small Bugs

Passing Ptr By Value

⚠️ Beware of Passing Smart Pointers by Value Passing a smart pointer (e.g. std::shared_ptr or boost::shared_ptr) by value is fine when you’re only modifying the object it points to. But if you int...

Posted by Rico's Nerd Cluster on February 1, 2025

Rico's Nerd Cluster