From a Multi-modal Intelligent Cell to a Self-organizing Robotic Skin - Realizing Self and Enriching Robot Tactile Interaction

Abstract

Human skin provides numerous inspirations for robots, supplying the whole body surface with multi-modal tactile sensitivity.Unlike a robot purely relying on joint information or vision, a robot equipped with artificial skin has a much richer information set.Challenges to efficiently deploy, organize and utilize a high number of distributed multi-modal sensors have so far prevented an effective utilization of artificial skin technology in robotics.In this thesis, we introduce a novel approach to create multi-modal artificial skin and a novel approach to self-organize the body representation of a robot.Our modular artificial skin is built by placing similar skin cells side-by-side into a flexible carrier material.Every skin cell is a self-contained system with a variety of sensors, signal conversion, processing and communication capabilities.The advantage of our modular approach is its robustness, scalability and transferability to various robotic systems.We developed various self-organizing features to automatically handle a potentially high number of skin cells on a large surface area.Automatic networking algorithms explore available skin cells and connections, distribute unique identifiers and provide robust and adaptive real-time communication.Mounted on a robot, our framework systematically explores and models the robot's body schema -inferring the robot's own kinematic and volumetric model from an egocentric perspective.In order to speed up the process, and to omit potentially harmful contacts, we only utilize low-range, open-loop motions of the robot and accelerometers embedded in our skin cells.A first algorithm explores the kinematic dependencies of body parts and joints, allocating actuators to joints and skin cells to body parts.A 3D reconstruction algorithm then computes the volumetric surface model of each body part, utilizing relative rotation estimates based on gravity and a topographic map inferred from the cell-2-cell connections.Turning skin patches into active visual markers, those distributed surface models can be visually combined into one homogeneous body representation -additionally joining tactile and visual space.A kinematic calibration algorithm finally estimates the parameters of the self-assembled kinematic model.In completion, we show exemplary applications of the prototype skin on industrial robot arms and the upper body of a humanoid robot.These examples demonstrate the benefits of an artificial skin for human robot interaction, multi-modal contact control, safety and object manipulation.