MIN Faculty Department of Informatics Bioinspired grasping in Soft Robotics University of Hamburg Faculty of Mathematics, Informatics and Natural Sciences Department of Informatics Technical Aspects of Multimodal Systems 11. November 2019 Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 1 / 30
Outline Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References 1. Motivation 2. Grasping: Definition & Basics 3. Charakteristics of Soft Robotics 4. Grasp Synthesis 5. Action-conditional model 6. Conclusion 7. References Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 2 / 30
Motivation Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Source: [1] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 3 / 30
Grasping - Problem Outline Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ N "fingers" on the grasping device ◮ ⇒ N contact points to the object ◮ How is the right Grasping Pose calculated? Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 4 / 30
Forces involved in grasping processes [3] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Normal force i w n Source:[2] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 5 / 30
Forces involved in grasping processes [3] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Normal force i w n ◮ Tangential force i w t Source: [2] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 6 / 30
Forces involved in grasping processes [3] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Normal force i w n ◮ Tangential force i w t ◮ Torsional moment i w θ Source:[4] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 7 / 30
Contacts [3] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Frictionless contact: i w n ◮ Frictional contact: i w n ∧ i w t ◮ Soft contact: i w n ∧ i w t ∧ i w θ Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 8 / 30
Equilibrium Grasp - Definition [3] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References A grasp is considered in equilibrium when: Wc + g = 0 , c � = 0 ◮ W := Wrench matrix ◮ c := Wrench intensity vector ◮ g := External wrench Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 9 / 30
Force-closed Grasp - Definition [3] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References A grasp is considered to be force-closed, when for every wrench ˆ w there is an λ that fits the constraints of a equilibrium grasp so that: W λ = ˆ w ◮ Note: Every force-closed grasp is a stable grasp Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 10 / 30
Categorization of Grasps Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Classes of grasps based on [3] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 11 / 30
Properties [3] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Properties of the grasp process: ◮ Dexterity ◮ Equilibrium ◮ Stability ◮ Dynamic behaviour ◮ Problems in grasping: ◮ Slipping detection ◮ Fracture of grasped object Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 12 / 30
Charakteristics of Soft Robotics [5] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Humanoid Soft Robotics ◮ Skeletton ◮ Metal ◮ Synthetic polymer ◮ Soft "skin" out of: ◮ Active elastomer ◮ Hydrogel ◮ Shape memory polymers ◮ e.g. GelSight, Dragon Skin, uSkin ◮ Animal-inspired Soft Robotics ◮ "CAN" be completely out of soft material Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 13 / 30
Charakteristics of Soft Robotics: Examples Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Source: [5] Source: [6] Source: [7] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 14 / 30
Grasp Synthesis [8] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Two kinds of approaches: ◮ Analytical ◮ Objective: Calculate possibly best configuration of position and angles ◮ Constrained optimization problems ◮ Based on 3D-models ◮ Data-driven ◮ Objective: Reusing existing grasp experience ◮ Heuristic ◮ Knowledge-based Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 15 / 30
Sensors [9] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Visual ◮ Depth sensing ◮ Pattern recognition ◮ ⇒ Stereo Camera Sensor ◮ Tactile ◮ Force sensing ◮ Surface exploration ◮ Slipping detection ◮ ⇒ GelSight, TacTip, etc. Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 16 / 30
Action-conditional model [10] Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References . . ◮ Objective: Combine visual and tactile sensing ◮ Sensors: GelSight tactile sensor, Microsoft Kinect v2.0 ◮ Operating with raw input data ◮ Self supervised Deep Learning approach to predict grasp success ◮ Adjusting grasps (Regrasping) ◮ Optimizable for gentle grasps Source: [10] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 17 / 30
2D tactile-sensor input Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Source: [10] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 18 / 30
Calculation of success probability Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Source: [10] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 19 / 30
Convolutional Neural Network Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Source: [11] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 20 / 30
Multi-layer Perceptron Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Source: [12] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 21 / 30
Regrasp Optimization Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Source: [10] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 22 / 30
Results Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Absolute sucessful Grasps Source: [10] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 23 / 30
Results Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Predicted success in relation to the applied force Source: [10] Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 24 / 30
Outlook Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References Multi-finger hand with uSkin: Source: [9] ◮ Recognizing objects based on tactile sensing with 95% success Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 25 / 30
Conclusion Motivation Grasping: Definition & Basics Charakteristics of Soft Robotics Grasp Synthesis Action-conditional model Conclusion References ◮ Soft Robotics are supportive for dynamic grasping tasks ◮ Visuo-tactile sensing is highly valuable for future grasping research ◮ But, more research is needed on: ◮ The combination of visual and tactile data ◮ Tactile sensors ◮ Suitable learning models for grasping Jan-Gerrit Habekost – Bioinspired grasping in Soft Robotics 26 / 30
Recommend
More recommend
