时间：2025年5月28日（周三）10:30-12:00

Time: 10:30-12:00, Wednesday, May 28, 2025

地点：云谷校区E10-404

Venue: E10-404, Yungu Campus

主持人: 工学院 赵世钰 博士

Host: Dr. Shiyu Zhao, School of Engineering

语言：中文

Language: Chinese

主讲人简介/Biography:

王越教授，浙江大学控制科学与工程学院。近五年来以通讯作者发表Nature Communication、IJRR、TPAMI、TRO、IJCV等期刊论文30多篇，发表RSS、ICRA、IROS等领域顶会30多篇，论文入选ESI高被引，获ICRA最佳视觉论文，IROS最佳设计论文提名等论文奖5次，入选年度全球前2%顶尖科学家榜单。担任期刊IEEE Robotics and Automation Letters编委、旗舰会议ICRA/IROS编委、IROS25本地主席、ARTS25本地主席等。获中国发明协会创业创新奖二等奖（排名1）等。主持科技部重点研发计划课题、国家自然科学基金、浙江省自然科学基金重大项目等.

讲座摘要/Abstract:

Over the past decade, the deployment of mobile robots in industry has seen remarkable progress, largely enabled by advancements in globally consistent Simultaneous Localization and Mapping (SLAM) systems, which serve as the backbone for robot localization and path planning. Nevertheless, critical challenges persist when envisioning robots operating autonomously over extended lifespans in human-centric environments like households. Persistent environmental dynamics and continuous streams of sensory data render the maintenance of globally consistent maps increasingly impractical. In this talk, we first present our efforts to address these challenges by improving the loop closure and state estimators that prioritize long-term robustness. Second, inspired by emerging paradigms in embodied AI for navigation, we critically rethink the necessity of global consistency in lifelong operation. By analyzing the complementary strengths of classical SLAM-based navigation and data-driven AI approaches, we try to bridge the gap and introduce a hybrid framework, and explore its performance in navigation benchmarks and dynamic environments.

西湖大学官网讲座信息链接：Rethink robot navigation in the AI era: Is global consistency necessary?