Meituan Unmanned Delivery Technical Salon – AI Research on Instance Segmentation, Visual Localization, Trajectory Prediction, and Depth‑Pose Learning

Event Time: Saturday, January 9, 2021, 14:00‑18:00

Location: 1st Floor, Building E, Wangjing Technology Park, Chaoyang District, Beijing

Registration: Click here to register

The rapid development of autonomous vehicles and aerial drones will profoundly affect daily life. Meituan, a leader in the life‑service industry, has been actively exploring unmanned delivery technologies. This technical salon presents the latest research achievements from Meituan’s unmanned delivery team, covering robot spatial positioning, instance segmentation, trajectory prediction, and planning control.

Speaker 1 – Mao Yinian, PhD Head of Meituan UAV Business Unit. Joined Meituan in 2018 after founding Beijing Erlangshen Technology. Recognized as a Beijing HaiJu Engineering expert and Chaoyang District Phoenix Plan technical expert. Holds a B.S. from Tsinghua University (2001) and a Ph.D. from University of Maryland (2006). Former Qualcomm Research engineer, author of 30+ patents and 20+ papers.

Topic 1: CenterMask – One‑Stage Instance Segmentation with Point Representation Speaker: Wang Yuqing, Algorithm Engineer, Meituan. Wang holds a master’s degree from Nankai University and focuses on object detection and instance segmentation for high‑precision map extraction. The talk introduces the CenterMask algorithm, originally presented at CVPR 2020. CenterMask: Single Shot Instance Segmentation with Point Representation

Topic 2: 3D Scene Geometry‑Aware Constraint for Camera Localization Speaker: Tian Mi, Algorithm Engineer, Meituan. Tian joined Meituan in 2018 and works on visual localization algorithms for unmanned delivery. The presentation covers a method published at ICRA 2020 that leverages 3D scene geometry constraints for accurate camera pose estimation. 3D Scene Geometry‑Aware Constraint for Camera Localization with Deep Learning

Topic 3: Multi‑Agent Trajectory Prediction with Global Interaction Speaker: Zhu Yanliang, Algorithm Engineer, Meituan. Zhu focuses on obstacle behavior analysis and prediction for unmanned vehicles. The talk describes the competition method “Robust Trajectory Forecasting for Multiple Intelligent Agents in Dynamic Scene” presented at ICRA 2020.

Topic 4: Attention‑Based Interaction‑Aware Spatio‑Temporal Graph Neural Network for Trajectory Prediction Speaker: Fan Mingyu, Researcher, Meituan. Fan is a research advisor in the unmanned delivery center, working on trajectory prediction and path planning. The method models dynamic obstacle interactions via an attention‑driven spatio‑temporal graph. Paper title: “An Attention‑based Interaction‑aware Spatio‑temporal Graph Neural Network for Trajectory Prediction”.

Topic 5: Online Baseline Calibration for Stereo Visual‑Inertial Odometry Speaker: Lang Xiaoming, Algorithm Engineer, Meituan. Lang develops visual positioning algorithms for the UAV division. The presentation introduces a real‑time stereo baseline calibration technique presented at ICRA 2020. Stereo Visual Inertial Odometry with Online Baseline Calibration

Topic 6: Self‑Supervised Depth‑Pose Learning in Dynamic Scenes Speaker: Gao Feng, Master Student, Tsinghua University. Gao’s research interests include robot localization, reinforcement learning, and multi‑robot coordination. The talk proposes a method that jointly learns scene motion and camera motion, generating adaptive supervision signals for dynamic environments. Attentional Separation‑and‑Aggregation Network for Self‑supervised Depth‑Pose Learning in Dynamic Scenes

Registration: Scan the QR code below for free registration.