Soft robotics with a focus on sensing.
Residual Physics for Grasp Failure Prediction
Prediction of grasping success is not a solved problem, with current research focusing on the grasp stability during lifting an object, which is much less then human intuition can do. Human intuition can assess the extent of possible movements, that can be done without losing a grasp of an object. The project attempts to produce an algorithm, which can analyze trajectory plans for a robotic arm and decide if the grasp would remain stable, based on tactile information, series of waypoints, and estimates of object mass and inertia. The chosen approach is to use residual physics, where a coarse physical model is complemented by residua l computed by a neural network. The project hopes to enable choosing optimal paths and maximum speeds for not optimal grasps.
Low-Cost Dextrous Robots for Food and Tool Handling
While the conventional robots are very successful in well-structures and predicable tasks and environments such as automobile assembling factories, they still underperform in unstructured and/or less-predictable tasks such as food preparation, cooking and associated tasks in domestic kitchens. To tackle these challenges, this project aims to develop soft robot manipulators that can perform some of these tasks to help simplify complex operations in kitchens in ordinary houses, possibly in cooperation with human users.
One of the main research drivers of this project is the use of sensorised soft robotic grippers that are able to handle variations of objects such as fruits, vegetables, plates, cooking tools etc. (water taps and dish washers) based on the guidance provided by computer vision. There are four main technological challenges in this domain as follows. First the closed-loop control of soft gripper interacting with a large variety of objects is a fundamental challenge. The use of mechanically adaptable structures needs to be utilised for grasping of a large variety of uncertain objects, while sensoring such systems needs a well-thought integration of soft tactile sensors into advanced feedback control processes, including the strategies to make the entire hardware setup reliable and economical. Second, the use of machine learning for visual recognition of variations of household objects is still a significant challenge particularly in unstructured environment and task, such as cleaning of dishes. And third, the interactions with human users in such an advanced platform are unsolved. What is the framework of human interfaces for complex robots for easy programming and teaching? How can humans give feedback to learning cooking robots? What is the framework of health and safety for such advanced robotic systems in household? The fourth is to address technological challenges allowing cost reduction in order to make robotic solutions more affordable and accessible. By developing this cutting-edge robotics platform in household, we will explore these fundamental questions in this project. The outcome of this project is also contributing to the automation of complex food manipulation tasks in the Agri-Food industry at large.