Topic：Robot Learning and Perception for manipulation and navigation

Date&Time：Dec.8th 7:00p.m.-8:30p.m.（Beijing Time）

Speaker：Prof. Ze Ji

From：Cardiff University

Voov：268 932 449（1836）

Abstract：

In this talk, he will share some of his work on reinforcement learning for robot manipulation, navigation, robot vision for navigation, and so on. Multistep manipulation tasks, such as block stacking or parts assembly, are complex for autonomous robots. A robotic system for such tasks would need to hierarchically combine motion control at a lower level and symbolic planning at a higher level. However, RL methods have limited capability to handle such complex tasks with many intermediate steps over a long time horizon. He will introduce the work that enables the agent to learn varied outcomes in multistep tasks. In addition, he will talk about some work on reinforcement learning for Mapless navigation, with human demonstrations and fail-safe localisation. Localisation and path planning are usually decoupled problems in robotics. His work introduces a simple and effective perception-aware navigation approach for more robust localisation and path following. Also, human demonstrations have been proved to be effective in accelerating the learning process. If time allows, he will introduce his work on machine learning for sensor self-diagnosis and self-recovery, where a deep learning-based approach is proposed to recover shock signals measured from a damaged high-g accelerometer, with self-diagnosis and self-calibration capabilities.

Introduction Of the speaker：

Dr Ze Ji a senior lecturer (associate professor) of Robotics and Autonomous Systems at the School of Engineering, Cardiff University, and Co-I of the centre for AI, robotics, and HMS (IROHMS). He received his PhD from Cardiff University, MSc in computer science from Birmingham University and BEng in Electronics Engineering from Jilin University. He has held multiple positions in both academia and industry (Dyson, Lenovo, and Autonomous Surface Vehicles), and gained rich project experience funded by EU FP6/FP7/H2020, Innovate UK, Royal Society, Industry (Renishaw, Spirent, Continental), etc. He has broad experience in robotics, including indoor mobile robots, industrial robot manipulators, and autonomous surface vehicles. His research is currently focused on robot vision, reinforcement learning, machine learning, simultaneous localisation and mapping (SLAM), tactile sensing, and their applications on autonomous robot navigation, manipulation and smart manufacturing.