http bdml stanford edu BioInspired Structures and Skin

http: //bdml. stanford. edu Bio-Inspired Structures and Skin for Human-Safe Robots M. R. Cutkosky Center for Design Research Stanford University 1

Toward human-safe robots: actuation + controls + skin & smart materials 2

What kind of skin and sensors should the next version of the arm have? • energy absorption • contact location • contact characterization 3

Human tactile sensing Meissner Corpuscle Merkel Disks Ruffini Organs Pacinian Corpuscle [graphic adapted from http: //psych. athabasca. ca/ html/Psych 402/Biotutorials/27/part 1. html] 4

Other biological sensors [1] • Slit sense organs – Scorpion: 76 / leg – Spider: 224 / leg 10 m [2] • Senses strains on the leg • Exoskeleton structures make the sensors possible • Able to sense forces applied to areas with no sensors (compared to FSRs) [1] F. G. Barth and M. Wadepuhl. Slit Sense Organs on the Scorpion Leg. Journal of Morphology, 145 (2): 209 -227, 1975 [2] F. G. Barth and J. Stagl. The Slit Sense Organs Arachids. Zoomorphologie, 86: 1 -23, 1976 5

Motivation for new approaches to design and fabrication • Traditional robot design… • …depends on stiff links, powerful actuators, numerous fasteners… Dante Robot • … assembled into a complex structure that does not tolerate unexpected events. Shaft coupling Shaft Motor Leg links Example linkage 6

Shape Deposition Manufacturing (SU/CMU) Embedded Component Part Support Deposit (part) Shape Deposit (support) Embed 7

Robot leg with embedded actuator, valves, sensor and circuitry Piston Leaf-spring spacer Valves Sensor and circuit Embedded components Detail of part just after inserting embedded components 7. Top support* 6. Part material 5. Embedded parts 4. Part material 3. Embedded sensor 2. Part material 1. Support material Sequence of geometries for fabrication Finished parts [Cham et al. 1999] 8

SDM: part number reduction, increased robustness, controlled compliance, damping Left: Kinematic prototype of linkage with 31 parts Center: SDM linkage with thick flexures, 1 part Right: SDM linkage with fabric-reinforced flexures 9

Example #2: mapping from passive mechanical properties to biomimetic robot structures Study biological materials, components, and their roles in locomotion. Study Shape Deposition Manufacturing (SDM) materials and components. viscoelastic material stiff material Models of material behavior and design rules for creating SDM structures with desired properties 10

Example #2: mapping from passive mechanical properties to biomimetic robot structures Study biological materials, components, and their roles in locomotion. 6 Force (m. N) 4 Data Model Study Shape Deposition Manufacturing (SDM) materials and components. Hysteresis loop @10 Hz 2 0 -2 -4 -6 -0. 5 -0. 4 -0. 3 -0. 2 -0. 10. 20. 30. 40. 5 position (mm) Models of material behavior and design rules for creating SDM structures with desired properties 11

Example #3: Stickybot Video S. Kim, Barrett Heynman 12

Example #4: Robot finger Joint 35 100 Shell 120 Fingertip Diameter: 35 Thickness of shell: 2 (unit: mm) New Exoskeleton Finger Existing Dexter Finger Hollow shell with embedded optical fibers and reinforced with metal mesh 13

Fiber Bragg Grating (FBG) Sensors 80 m dia. • Light, small, flexible and easy to embed • Electrically passive • Can optically multiplex Optical Fiber Input • High resolution & accuracy Transmission Reflection • Physically robust FBG S 1 S 5 S 2 Strain Sensor S 4 Temperature Compensation Sensor S 5 S 3 Ph. D students: Y-L Park Collaborators: IFOS Inc. 14 Funding: NASA, IFOS

Sensor Configuration • Strains are mostly concentrated on the top of the shell • Four strain sensors at 90 interval ( first rib of the shell ) • One temperature compensation sensor in the middle S 1 S 5 S 2 Strain Sensor S 4 Temperature Compensation Sensor S 5 S 3 15

Applying SDM to the Human-Safe Robot • Compliant protective Layer • Pressure/Force sensing • Proximity sensing • ~1 cm spatial resolution • ~0. 03 N minimum force • ~100 Hz sample rate 16

Tactile sensing technologies* Many transduction technologies and sensor designs… – – – piezoresistive capacitive optical quantum tunneling piezoelectric. . . Issues: – resolution, dynamic range – robustness – ease of integration into skin – ease of multiplexing and connecting – noise immunity – hysteresis *Cutkosky, Howe, Provancher, “Cha. 19 Force and Tactile Sensors, ” in Springer Handbook of Robotics, Siciliano & Khatib, eds. , 2008 17

Capacitive Sensing Proximity Transduction Measurement + + + –– ++–+– + – + +––– –– ΔQ 18

Capaciflector* Electric Field Sensitive Region (Thickness indicates sensitivity to objects) Sensing Plate Large contribution to capacitance Shielding Plate Ground Sensing. Plate Ground Plate Shielded Region *NASA and Stanford/VA (N. Smaby) 19

Dual Mode Capaciflector Electric Field (Thickness indicates sensitivity ) Proximity Sensitive Region Stiff Material Proximity Sensing Plate Pressure Sensing Plate Ground Plate Pressure Sensitive Region Shielded Region Soft Material Barrett Heyneman 20

Dual Slope: Conversion VREF VOUT time CS V+ RX CX Repeat + RD VREF VOUT Charge sensing capacitor, CX Transfer charge to storage capacitor, CS Discharge CS at known rate through RD 21

Dual Slope: Measurement CS VREF V+ VOUT RX CX Slope: VREF/RDCS Known # of steps: n time Process + RD VOUT VREF Variable, measured time: t 1. Measure charge, QS, transferred to CS a. Measure time t b. QS = t*VREF*RD 2. Determine charge, QX, per step a. QX = QS/n 3. From switching time and QX calculate CX Trade-off Want: High resolution and low total time a. Fixed time resolution Δt fixes QS resolution, ΔQ = Δt*VREF*RD b. Variable time t = RD*CS c. => Increasing RD increases both resolution and time 22

Tri-Slope: Conversion VREF VOUT time CS V+ RX VREF CX RD CD V+ + VOUT Repeat Charge sensing capacitor, CX Transfer charge to storage capacitor, CS Transfer charge from CS to Fine Coarse Repeat drain capacitor, CD Discharge CD 23

Tri-Slope: Measurement CS V+ VREF VOUT RX VREF + VOUT CX Known # of steps: n time Process Variable, measured # of coarse, fine steps: NC, NF CD RD V+ Benefits 1. Measure charge, QS, transferred to CS 1. High Accuracy a. Measure number of coarse/fine a. QF determines QS measurement steps: NC, NF resolution b. Charge per course/fine step: QC, QF 2. Fast Measurement c. QS = QC*NC + QF*NF a. Coarse steps quickly drain CS a. Determine charge, QX, per step 3. QC/QF can be adjusted for high accuracy, a. QX = QS/n fast, and roughly constant time b. From switching time and QX calculate CX measurements 24

Frequency Mode Capacitive Sensing – The Sensor A simple circuit with small, low-cost components is replicated for each sensor. Concept: Schmitt inverter, resistor, capacitor ==> oscillating output. Electrical noise is virtually eliminated by placing sensing circuits next to the sensors and using shielding. multivibrator and switching The shield is part of the sensor! John Ulmen 25

Frequency Mode Capacitive Sensing – Array Format • Excellent noise • Highly scalable immunity • ≈ $0. 20/sensor • Digital pulse output • Row column sensor addressing • Single output wire 26

Optical Sensing Emitter • Rugged, flexible, transparent materials. • Relies on internal reflection of infrared light (immune to electromagnetic interference). • Geometry, materials, influence sensitivity. • Can tile sheets for covering large areas Sensor Undisturbed In Compression In Bending Dan Aukes Cross section of sheet showing effects of compression and bending 27

Two-dimensional optical prototype • An optically clear elastomer is molded in a thin sheet. • Rows of infrared-light-emitting emitters and detectors are embedded in the material. • Light traveling through the material is diminished as forces are applied. • A two-dimensional pressure profile can be created by combining information from multiple emitters and detectors. • Multi-touch sensing is possible. 28

Electrical Design • Emitters are turned on one at a time. • An analog reading is taken at every detector for each emitter. • Number of (useful) readings: 32 • Number of detectors: 8 realtime display utility that communicates with PIC microprocessor 29

Next Steps Optical and capacitive: • • • Build arrays and test sensitivity when mounted on compliant arm covering. Develop flexible printed circuits (reduce wiring). Develop bus and communications protocol. Test energy absorption, robustness. Investigate sampling rate/accuracy tradeoff. Capacitive: • • • Develop sensor array Evaluate proximity and contact sensing accuracy. Test sensitivity to noise and stray capacitances. Optical: • • • Optimize geometry and materials of flexible sheet to improve sensitivity. Separate contact FIR effects from bending. Develop solution for tiling with minimum “dead area” at emitters, detectors. 30

Materials • Clear. Flex 50 – – – • Rubber Glass II – – – • Polyurethane rubber Quite transparent Index of refraction ~ 1. 5. Tear resistant Stiff Bonds to other rubbers Silicone rubber extremely brittle more transparent than Clear. Flex, softer than clear flex, with less mechanical hysteresis cannot be coated or painted, Vyta. Flex 10 -60 – – – polyurethane rubber not completely transparent. better mechanical properties than Clear. Flex. Can bond with clear-flex, making it well-suited for future multi-material designs. 31

Specifications • • • Sensitivity – the sensor should be able to sense between 10 g-10 kg of weight. Location – the sensor should be able to localize forces to within 1 cm. Mounting location – the sensor should be able to perform all its sensing tasks on a non-rigid, non-continuous surface (such as a metal or plastic mesh or screen). Flexible – the sensor should be able to bend without breaking. Conformable – the sensor should be able to take the shape of the surface it adheres to. Connectable - rigid wiring schemes should not interfere with the flexibility of the skin. Protective - The skin should introduce a layer of safety between the robot and environment. Lightweight - The skin's weight should not severely impede performance. Processing – The information that comes from the skin needs to be able to be processed in an efficient manner without losing too much information. Frequency – sensor data should be refreshed at a frequency of at least 1 k. Hz 32