Rico's Nerd Cluster

「离开世界之前 一切都是过程」

[BEV] Feature Pyramid Network

FPN = Feature Pyramid Network

[BEV] BEV Ideas and View Transformation

Training pipelines, multiview geometry, and how to lift image features into BEV

1. The Core Goal of a BEV Network A Bird’s Eye View (BEV) network aims to reconstruct, in real time using only onboard cameras, a local 3D map of comparable quality to what an offline reconstructi...

[BEV] BEV Introduction: Tesla's Challenges and Architecture

Why per-camera detection falls short, and how BEV solves it

1. Tesla’s Perception Challenges (2021) Before BEV, Tesla’s pipeline detected objects and lanes independently in each camera view and then tried to fuse the results. This created fundamental probl...

[BEV] BEV Introduction

Understanding L2 and L2+ Perception: Why the Real Challenge Is System Design, Not Just Better Detection When people talk about assisted driving, they often jump straight to compute power, percepti...

MPPI And Motion Control On My Diff Drive

Remote Claude Code Setup

Motor Control So, tested on carpet, I think it’s about not having (+/-) on the spot that makes it motor controller - aware that the robot turns better using something like: left = ±u righ...

AI Agent Setup

Remote Claude Code Setup

Remote Claude Code Setup Bring up tailscale: sudo tailscale up Bring up code-server: code-server --bind-addr $(tailscale ip -4):8081 To see password for access cat ~/.config/code-server/con...

Computer Vision - Charuco Board and Phone Pose Detection

Mission The goal is to generate an object ground-truth pose using an RGB-D camera, a ChArUco board, and a known object CAD model or object dimensions. The ChArUco board gives us a stable metric c...

Computer Vision - Charuco Board Detection

ChArUco Pose Estimation A ChArUco board is basically a chessboard combined with ArUco markers. An ArUco marker is a square fiducial marker with a binary ID pattern inside it: 1 2 3 4 5 6 marker ...

Computer Vision - Aruco Detection

Aruco Detection Aruco detection has three jobs: 1 2 3 1. Detection: find square marker candidates in the image. 2. Decoding: read the binary ID inside the square. 3. Pose estimation: use the four...

Computer Vision - Pose Estimate Pipeline

Building a 2D-to-6D Object Pose Pipeline with RF-DETR and FoundationPose I built a small pose-estimation pipeline that combines RF-DETR and FoundationPose. RF-DETR solves the 2D detection problem:...