Robotica Umanoide. Lorenzo Natale icub Facility Istituto Italiano di Tecnologia. 30 Novembre 2015, Milano

Transcription

1 Robotica Umanoide Lorenzo Natale icub Facility Istituto Italiano di Tecnologia 30 Novembre 2015, Milano

2 Italy Genova

3 Genova Italian Institute of Technology Italy

4 Genova Italian Institute of Technology Italy IIT icub Facility

5 Industrial robots, assembly line: Fast, reliable, precise, but highly specific Robotics, now

6 Industrial robots, assembly line: Fast, reliable, precise, but highly specific Need big effort to engineer the environment Everything is preprogrammed, little or no sensing

7 no humans!

8 Robotics, now Roomba, the robot cleaner, more general but it does not do much!

9 New robots: applications Space environment/household applications (Justin, Valkyrie) Household applications (PR2, Romeo, Pepper) Military (Boston dynamics) Civil (Walkman, Atlas, Nimbro) Industrial applications (Rethink robotics, Redwood robotics, SMErobotics) Research platforms (icub, PR2, HRP, Asimo, Meka robots, Rethink robotics)

10

11

12 Robots in Research Labs can be quite successful ATR, Japan

13 ...In the real world

14 But why are robots still stuck in research labs?

15 Simple problems for computers In 1997 Deep Blue won a six game match against world champion Kasparov Massive parallel computation, able to generate and evaluate 200M positions per seconds

16 Simple problems for humans What a machine sees What we see Adapted from: P. Fitzpatrick, et al. 2008, original work from Martin et al. 2004

17 Learning from many examples Thousands of images of faces (face detection) Thousands of examples of a class (8k types of cats)

18 Embodied intelligence Elephants don t play chess! (Brooks 1990) The world is its own best model, the trick is to sense it appropriately and often enough The essence of intelligence the ability to move around in a dynamic environment, sensing the surroundings to a degree sufficient to achieve the necessary maintenance of life and reproduction Full video:

19

20

21 The real world: problems So called unstructured environment Largely unmodeled Unexpected interaction is normal: with the environment and with humans Safety

22 Research topics and outline interaction objects platform system integration tools Engineering Research/science

23 Many platforms, many solutions Lots of different robots, scenarios and experimental settings

24 Giorgio Metta (integration) Chiara Bartolozzi (avls) Lorenzo Natale (sensing & software) Francesco Nori (control)

25 Shared open platforms Different platforms and experimental scenario makes it difficult to compare experiments Slow incremental progress Reinventing the wheel Lacks of standards Open software Shared platforms

26 Shared platforms Open software Read code Customize and build upon, share results Shared platforms: Make robots available to non roboticists Community: share code, compare experiments Robots: icub, PR2, Nao, Darwin-Op

27 Why humanoids

28 The icub Open platform (software and hardware) Community: 30 copies built

29 icub Evolution

30 Why is the icub so special? hands: we started the design from the hands 5 fingers, 9 degrees of freedom, 19 joints sensors: human-like, e.g. no lasers cameras, microphones, gyros, encoders, force, tactile electronics: flexibility for research custom electronics, small, programmable (DSPs) reproducible platform: community designed reproducible & maintainable yet evolvable platform large software repository (>1M lines of code)

31

32 Research and engineering perception, control and learning brain vision In collaboration with our industrial partner new actuation force/torque sensors energy computation new tools skin-like sensors

33 can bus/ethernet Brain Ethernet MTB 6SG 2FOC xyz Distributed cluster Local network

34 Software platform Open source community (116 members) Using the software without robots: dataset player, simulators (gazebo, robotran) Models for icub, Coman, Walkman etc Available with installer for Windows and Linux

35 Control of the interaction

36 Controlling the interaction Industrial robots Position controlled, do their best to go to a certain position and stay there: stiff Not suitable for physical interaction Interaction if expected must happen at the end-effector (hand) Little or no control (no brain) Little constraints on design: huge and heavy Dangerous

37 Research robots Interaction can happen everywhere Control the amount of force exerted Autonomous (actuation, energy and computation) Light Safe

38 Research robots Interaction can happen everywhere Control the amount of force exerted Autonomous (actuation, energy and computation) Light Safe

39 Safety and robustness DLR German Aerospace Center,

40 Research challenge Build robots that can control the amount of force exerted on the environment Force control Impedance control: spring like mechanism with variable stiffness software: elasticity is simulated (easier but less effective) mechanical: elasticity is intrinsic (better but difficult)

41 An example of elasticity (simulated)

42 Intrinsic elasticity Credits: M. Laffranchi, Advanced Robotics, IIT

43 Credits: Advanced Robotics, IIT

44 Tactile sensors

45 icub tactile system Tactile sensing is a key technology for robots: Safety, control of interaction forces, human robot interaction Widely studied: mostly transduction principles and technologies Need for tactile sensor system which can detect contact over large areas easy integration: reduce number of wires, deformable, cheap easy to replace, modular

46 capacitor Main idea: build a soft capacitor ground plane: e.g. conductive fabric parameters: mechanical properties, impedance, etc. soft material: e.g. silicone parameters: dielectric constant, mechanical stiffness, etc. electrodes: etched on a flexible PCB parameters: shape, folding, etc.

47 Basic element Interconnected triangular elements (12 sensors) CDC chips on each triangle, performs capacity-to-digital conversion I2C interface

48 A mesh of sensors

49 External layer Layers of fabrics glued together with the Lycra in a sandwich Elastic deformable fabric replaces silicone foam (dielectric) Design and fabrication based on techniques adopted in clothing industry protective Lycra conductive Lycra Soft Dielectric 3D fabric

50 icub Sensorization Hands: 104x2 Forearms: 230x2 Upperarms: 380x2 Torso: 440 Legs and feet: 1310x2 Total: accelerometers in the palms and arms

51 Integration on different robots

52 Video: contact force estimation and compensation

53 Whole-body control

54 Fingertips: special solution with same principle 12 sensors 14.5 x 13 mm Fabric, conductive+protective Plastic interface transforming to curved shape Inner support for PCB PCB fingernail CDC chip

55 Object manipulation

56

57 Human robot interaction

58 Human robot interaction New application domains implies closer interaction between humans and robots Program robots with easy and intuitive interface so that they can better understand humans and interact naturally with them Ability to recognize speech, gestures and facial expressions Shared goals, proactive behavior Beneficial for robots: teach by demonstration, human as caregiver (see later) It requires safety

59 H. Ishiguro, ATR, Japan the uncanny valley!

60 Nice robots Should be careful not to attempt to mimic humans too closely Gaze Smooth motion Natural interaction: speech, vision, touch

61 Meka S2 Humanoid Head

62 Natural behavior

63 The icub Talking Head

64

65 Teaching by demonstration

66 Teaching by demonstration P. Kormushev, S. Calinon, D. Caldwell, Advanced Robotics, IIT

67 Visual perception (objects and tools)

68 Visual Perception Depth map Segmentation Detection of humans Motion computation & egomotion Objects

69 Object perception

70

71 Learning with human supervision

72 Tools

73 Towards tool use...

74 Exploring tool size

75 Learning to Use Tools (1)

76 Learning to Use Tools (2)

77 Software engineering

78 Generalize objects based on their use

79 other experiments

80 Acknowledgements Giulia Pasquale Tanis Mar Ali Paikan Massimo Regoli Nawid Jamali Carlo Ciliberto Alessandro Roncone Vadim Tikhanoff Ugo Pattacini Lorenzo Rosasco Giorgio Metta