Contact-rich Manipulation & Cooking Robot Mastery - Dr Joao Moura and Dr Namiko Saito (University of Edinburgh)

The Statistical Machine Learning and Motor Control Group of the University of Edinburgh.
The Statistical Machine Learning and Motor Control Group of the University of Edinburgh.

Date: 17 June 2024 11:00 - 12:00

Location: G27, Engineering building (west block, ground floor), Mile end campus

This is a double robotics seminar with two speakers from the Statistical Machine Learning and Motor Control Group of the University of Edinburgh: Dr Joao Moura and Dr Namiko Saito. The general topic is robotic manipulation using machine learning and control techniques, including applications to robotic cooking! More detailed information below.

Contact-rich Manipulation --- Bridging the gap between planning and control (Dr Joao Moura)

Abstract: Contact-rich manipulation refers to manipulation scenarios requiring contact and force interactions with the environment. Examples of such tasks include wiping a surface and non-prehensile manipulation tasks, such as, pushing or catching objects. Contacts interactions with the environment introduce hybrid dynamics and additional uncertainty from friction and/or impacts, which is challenging to model and control. In this talk, I will overview some of the planning and control approaches we have been developing for achieving the above dynamic tasks with physical robotic manipulators. In particular I will discuss the role and interplay of prediction, for motion generation, and feedback, for motion control.

Profile: Joao is a research associate at the University of Edinburgh in the Statistical Machine Learning and Motor Control Group (SLMC), affiliated to the Alan Turing Institute. He has Ph.D. in Robotics and Autonomous Systems, jointly awarded by Heriot-Watt University and The University of Edinburgh, in 2021, and an Integrated Master's degree in Mechanical Engineering from The University of Aveiro, Portugal, in 2016. His research interests include contact-rich and non-prehensile manipulation, trajectory optimization, model predictive control, and imitation learning.

Cooking Robot Mastery: Learning Online Motion Generation with Active Perception (Dr Namiko Saito)

Abstract: In this presentation, I introduce a predictive recurrent neural network with an attention mechanism that can weigh the sensor input, distinguishing how important and reliable each modality is, that realises quick and efficient perception and motion generation. We tackled the task of cooking scrambled eggs using real ingredients by a real humanoid robot, in which the robot needs to perceive the states of the egg and adjust stirring movement in real time, while the egg is heated and the state changes continuously. Although handling changing objects is challenging because sensory information includes dynamical, both important or noisy information, and the modality which should be focused on changes every time, we solve it with our deep neural network model which can conduct both perception and motion generation in real time.

Profile: Namiko received the B.S., M.S., and Ph.D degrees in mechanical engineering from Waseda University, Japan in 2018, 2020 and 2022 respectively. Since 2023, she has been a Research Associate in The University of Edinburgh, affiliated with the Alan Turing Institute and a visiting researcher at Waseda University. Her research interests include multimodal learning, sensorimotor control, and dexterous manipulation.

Contact:  Lorenzo Jamone

Updated by: Lorenzo Jamone