ROBOTIC INTELLIGENT ASSEMBLY（机器人智能装配）

本书致力于介绍机器人智能装配领域的*新进展，重点聚焦于轴孔装配（PiH）任务。书中涵盖了机器人轴孔装配策略的多个维度，包括感知、基于模型的控制与基于学习的控制。作者系统梳理了以往的研究成果，帮助读者全面理解该领域。

This book explores peg-in-hole assembly strategies to study robotic intelligent assembly.It presents several state-of-the-art principles for peg-in-hole assembly strategies,supported by experimental evidence.In pursuit of theoretical innovation,the author summarizes their research on robotic intelligent assembly over the past decade,highlighting the limitations of model-based algorithms in complex assembly environments and the importance of data efficiency for learning-based algorithms.Each algorithm is supported by extensive experimentation and results demonstrating its effectiveness.A review of research ideas provides readers with a comprehensive understanding of the progress made in this field.This monograph is intended for undergraduate and postgraduate students interested in robotic intelligent assembly,researchers studying robotic intelligent assembly algorithms,and electronic,mechanical,and computer engineers engaged in industrial robot-assisted assembly.

徐静，清华大学长聘教授、博士生导师。研究领域：具身智能、机器人视觉和触觉、机器人智能操作、仿生机器人、智能制造围绕智能机器人与智能制造，先后承担了国家杰出青年科学基金项目等项目三十余项，发表SCI期刊论文100余篇，多篇论文入选高被引论文；授权发明专利60余项，以第一完成人获得北京市技术发明奖一等奖等多个奖励。

1 Introduction

1.1 Background

1.2 Robotic Peg-in-Hole Assembly System

1.2.1 Manipulators

1.2.2 Mating Parts

1.2.3 Sensing

1.2.4 Assembly Strategies

1.3 Literature Review

1.3.1 Multi-View Stereo

1.3.2 Tactile Sensor Simulation

1.3.3 Model-Based Strategies

1.3.4 Learning from Demonstrations

1.3.5 Learning from Environments

1.4 Scientific Goals

1.5 Book Organization

References

2 Multi-View Stereo Sensing Method

2.1 Introduction

2.2 Method

2.2.1 Visibility-Aware Feature Aggregation

2.2.2 Coarse Depth Prediction

2.2.3 2D-3D Feature Lifting

2.2.4 PointFlow

2.2.5 Training Loss

2.3 Experiments

2.3.1 Experiments on Synthetic Data

2.3.2 Experiments on DTU Benchmark

2.3.3 Network Design Analysis

2.3.4 Generalizability of the PointFlow Module

2.4 Conclusion

……