A Deep Learning Method for Vision Based Force Prediction of a Soft Fin Ray Gripper Using Simulation Data.

Soft robotic grippers are increasingly desired in applications that involve grasping of complex and deformable objects. However, their flexible nature and non-linear dynamics makes the modelling and control difficult. Numerical techniques such as Finite Element Analysis (FEA) present an accurate way of modelling complex deformations. However, FEA approaches are computationally expensive and consequently challenging to employ for real-time control tasks. Existing analytical techniques simplify the modelling by approximating the deformed gripper geometry. Although this approach is less computationally demanding, it is limited in design scope and can lead to larger estimation errors. In this paper, we present a learning based framework that is able to predict contact forces as well as stress distribution from soft Fin Ray Effect (FRE) finger images in real-time. These images are used to learn internal representations for deformations using a deep neural encoder, which are further decoded to contact forces and stress maps using separate branches. The entire network is jointly learned in an end-to-end fashion. In order to address the challenge of having sufficient labelled data for training, we employ FEA to generate simulated images to supervise our framework. This leads to an accurate prediction, faster inference and availability of large and diverse data for better generalisability. Furthermore, our approach is able to predict a detailed stress distribution that can guide grasp planning, which would be particularly useful for delicate objects. Our proposed approach is validated by comparing the predicted contact forces to the computed ground-truth forces from FEA as well as real force sensor. We rigorously evaluate the performance of our approach under variations in contact point, object material, object shape, viewing angle, and level of occlusion.

A Passive Learning Sensor Architecture for Multimodal Image Labeling: An Application for Social Robots.

Object detection and classification have countless applications in human-robot interacting systems. It is a necessary skill for autonomous robots that perform tasks in household scenarios. Despite the great advances in deep learning and computer vision, social robots performing non-trivial tasks usually spend most of their time finding and modeling objects. Working in real scenarios means dealing with constant environment changes and relatively low-quality sensor data due to the distance at which objects are often found. Ambient intelligence systems equipped with different sensors can also benefit from the ability to find objects, enabling them to inform humans about their location. For these applications to succeed, systems need to detect the objects that may potentially contain other objects, working with relatively low-resolution sensor data. A passive learning architecture for sensors has been designed in order to take advantage of multimodal information, obtained using an RGB-D camera and trained semantic language models. The main contribution of the architecture lies in the improvement of the performance of the sensor under conditions of low resolution and high light variations using a combination of image labeling and word semantics. The tests performed on each of the stages of the architecture compare this solution with current research labeling techniques for the application of an autonomous social robot working in an apartment. The results obtained demonstrate that the proposed sensor architecture outperforms state-of-the-art approaches.