Public defence in Automation, Systems and Control Engineering, M.Sc. David Blanco-Mulero

The title of the thesis: Towards Efficient Robotic Manipulation of Deformable Objects by Learning Dynamics Models and Adaptive Policies

Doctoral student: David Blanco-Mulero
Opponent: Prof. Yiannis Demiris, Imperial College London, United Kingdom
Custos: Prof. Ville Kyrki, Aalto University School of Electrical Engineering, Department of Electrical Engineering and Automation

To deal with the global growth of the ageing population, intelligent robot systems hold the promise to assist humans in daily life tasks such as assisted dressing or feeding. Thanks to recent advances in learning-based approaches, robotic systems have been able to successfully manipulate a wide variety of objects. However, most of this success has been largely focused on rigid objects, overlooking the deformable objects encountered in household chores, such as textiles or vegetables. These objects introduce a new set of challenges, such as representing their deformation and adapting the robot manipulation actions accordingly. 

To address this problem, this dissertation proposes methods to bridge the gap in the adaptive capabilities of robotic systems for manipulating a variety of materials and objects. More specifically, it investigates methods that can learn to efficiently manipulate deformable objects in simulation, transfer the learned skills to the real world, and examine the challenges that arise when transferring these skills. First, the dissertation explores the representation and modelling of deformable object dynamics using data-driven approaches. Then, it proposes methods for learning manipulation policies that can adapt and generalise to deformable objects varying in shape, size and material properties. Finally, it investigates a major challenge in learning approaches that leverage simulation engines, the sim-to-real gap, by quantifying the reality gap in cloth manipulation tasks. 

Together, the results of this dissertation indicate that policies learnt in simulation can adapt to a wide variety of deformable objects and can efficiently manipulate them, where closed-loop feedback can mitigate the reality gap in these approaches. Consequently, approaches based on learning in simulation can enhance the adaptability of manipulation systems, where closed-loop feedback plays a vital role in successfully transferring the learnt skills to the real world.

Thesis available for public display 10 days prior to the defence at: https://aaltodoc.aalto.fi/doc_public/eonly/riiputus/