Disentangling perception and reasoning for improving data efficiency in learning cloth manipulation without demonstrations

Cloth manipulation is a ubiquitous task in everyday life, but it remains an open challenge for robotics. The difficulties in developing cloth manipulation policies are attributed to the high-dimensional state space, complex dynamics, and high propensity to self-occlusion exhibited by fabrics. As analytical methods have not been able to provide robust and general manipulation policies, reinforcement learning (RL) is considered a promising approach to these problems. However, to address the large state space and complex dynamics, data-based methods usually rely on large models and long training times. The resulting computational cost significantly hampers the development and adoption of these methods. Additionally, due to the challenge of robust state estimation, garment manipulation policies often adopt an end-to-end learning approach with workspace images as input. While this approach enables a conceptually straightforward sim-to-real transfer via real-world fine-tuning, it also incurs a significant computational cost by training agents on a highly lossy representation of the environment state. This paper questions this common design choice by exploring an efficient and modular approach to RL for cloth manipulation. We show that, through careful design choices, model size and training time can be significantly reduced when learning in simulation. Furthermore, we demonstrate how the resulting simulation-trained model can be transferred to the real world. We evaluate our approach on the SoftGym benchmark and achieve significant performance improvements over available baselines on our task, while using a substantially smaller model.

GARField: Addressing the visual Sim-to-Real gap in garment manipulation with mesh-attached radiance fields

While humans intuitively manipulate garments and other textiles items swiftly and accurately, it is a significant challenge for robots. A factor crucial to the human performance is the ability to imagine, a priori, the intended result of the manipulation intents and hence develop predictions on the garment pose. This allows us to plan from highly obstructed states, adapt our plans as we collect more information and react swiftly to unforeseen circumstances. Robots, on the other hand, struggle to establish such intuitions and form tight links between plans and observations. This can be attributed in part to the high cost of obtaining densely labelled data for textile manipulation, both in quality and quantity. The problem of data collection is a long standing issue in data-based approaches to garment manipulation. Currently, the generation of high quality and labelled garment manipulation data is mainly attempted through advanced data capture procedures that create simplified state estimations from real-world observations. In this work, however, we propose to generate real-world observations from given object states. To achieve this, we present GARField (Garment Attached Radiance Field) a differentiable rendering architecture allowing data generation from simulated states stored as triangle meshes. Code will be available on https://ddonatien.github.io/garfield-website/