Inheriting Humanoid Systems from Remote-Brained Robotics

Transcription

1 Inheriting Humanoid Systems from Remote-Brained Robotics Masayuki Inaba Professor, JSK Robotics Lab. Department of Mechano-Informatics/ Department of Creative Informatics University of Tokyo 1 Prof., 1 Associate Prof. 3 Assistant Profs., 35 students, 2013FY 1

2 How? Towards Whole-Body Humanoid Development Methodology Software Hardware Body Type 1983 Rope Handling 1 st ISRR 2013 Dancing Steps 1988 Dexterous Manipulation Environment Behaviors Task Integration Inheritance 2013 Musculoskeletal Humanoid

3 Remote-Brained Robotics Mother Environment th ISRR Embedded Brain Agent Brain Mother Environment Embodied Agent Sensor s Motors Interface Body Physical Environment Concurrent Development Remote Brained Remote-Brained 2000 Robotics: about 60 robots

4 Body Brain Common Wireless Interface 1993 Wireless receiver transmitter RC servos M. Inaba: Remote-Brained Robotics: Interfacing AI with Real World Behaviors, in Robotics Research: The Sixth International Symposium on Robotics Research (6 th ISRR), pp , International Foundation for Robotics Research,

5 Software Abstraction Platform for Different Type Bodies 1993 ISRR Mother Wireless Tower Multi-robot Brain Internet Dual Mobile arms Hetero robot Quadruped Apelike Body Humanoid 5

6 Shared Mother Environment for different brain-body combinations Study on Robot Cooperative Tasks such as Handover 1998

7 Vision Based Whole Body Motion 1993 Typical Task: Swinging with different brains 1) Human solved controller 2) Learning controller Tracking Vision

8 From Swing to Brachiation 1997 Swing and reach vision and grasp sensors

9 Behavior Developmental Research Knuckle walk 1993 Balancing 1995 Balance Manipulate 1997 Rollover 1994 Learning 1995

10 Body Sensor Image in Brain Tactile Sensor Suit th ISRR Tactile Signal area Integrated Sensor Image Electrically Conductive Fabric and String

11 1992 JSK, 1989 JSK, software on 48 Trasputers Single Trasputer with a MEP Stereo and 3D Tracking Interaction Navigation

12 Study on connection from perception to action AI: Perfect Planning Model Feed-forward R: Attention Selection Feedback loop B: Motion Arbitration Parallel Structure Learning Learning Learning Robotics is the study of intelligent connection from perception to action st ISRR

13 Single Proces s Model Perception Action If/then simple control Simple Programmed Behaviors Parallel Sensor Monitoring Process

14 Parallel Processes from P to A Perception Action Parallel Evaluation of Sensor States Action Arbitration

15 Process Network from Perception to Action Perception Action Data-driven control, extensible with modularity Behavior Network (BeNet)

16 Process Network for Error Recovery Using State Network Upper layer: State Net Lower layer: Behavior Net

17 Learning connection between Perception to Action GA Virtual in Mother Controller parameters

18 From small-size to life-size Workstation to PC PC becomes smaller and powerful 1995 Auto-balancer by Small 1999 Auto-balancer by H5 Software Extensible Software RT Servo Controller 2001 ladder by Small humanoid 2004 ladder by Life-size humanoid H7

19 Life-size Humanoid to common Platform H6 & H7 Humanoid Dual CPU PC board Hardware Maintena nce by Company HRP2 Project Stereo Vision on board PC Correlation based Software Implementation with MMX HARP: Humanoid Autonomous Robot Project H1-H

20 HRP2 Tool manipulation kettle pouring water dust cloth wiping broom sweeping cleaner operation 20 playing instrument washing machine operation

21 From Humanoid to Specialized Robots in Industry Collaboration Projects 2006 Audio Visual Care Robot With Fujitsu Dexterous Sensor-based Manipulation With Panasonic Common Integrated Software for Perception Manipulation Robust Task Sequence Execution With Toyota IRT

22 High Speed High Powered Motor Driver 2010 Toward DRC 2012 SCHAFT Liquid cooled motor Dev

23 New fabrication of body with free-shape bones 1995 Musculoskeletal Humanoid

24 Musculoskeletal Humanoid More than 100 motors with wire tension driven robot. Musculoskeletal Humanoid

25 Total System and Mother JSK Environment Real Body ( 技 術 解 説 資 料 ) Desktop Type Mother Environment Task model translator Higher Layer Interaction Control Behavior Statenet Localization Navigation Motion Planning Body Model Desktop Type Furniture Type IRT 基 盤 ソフトウェア Musculo Memory Task Modeling Success Conditions Env. Body Modeling Furniture Musculoskeletal Mobile Manip Personal Mobility Mobile Manip PM Environment Modeling Humanoid Realtime Reactive Layer Body Monitoring Reactive Control Sensor-Motor Body I/O Programming Window

26 All JSK Robots in Euslisp Abstracted description of humanoid robots Joint space level commands (send *kaz3* :larm :shoulderp :joint-angle 80) Work space level commands (setq c (send *kaz3* :larm :endcoords :copy-worldcoords)) (send *kaz3* :larm :inversekinematics c) robot-frame robot-limb robot-limb robot-limb robot-link robot-link robot-link joint COSMOS: Lisp-based Robot System Integration Environment

27 Simulation example on EusDyna 2008 Whole body effect Gripper Standup and walking Multi contacts

28 With OSS Community Open Source ROS Language from OMICS (Open Mind Indoor Common Sense) TUM collaboration Building Map Saito, Chen, Okada, Inaba, Kunze, Beetz, IROS 2011

29 Remote Virtual Challenge and Lab OSRF Cloud Sim, Virtual DRC Simulator on remote cloud system at US HRP2 Experiments in Cloud Room, 2013 PointCloud, Camera data taken at Tokyo HRP2 Experiments in U-Tokyo Room, 2006

30 Remote Testing Server tum cmu aist gt jsk freiburg Cloud Jenkins Server Download Source Continuous Integration Tool: Jenkins ROS, OpenHRP, RTM: Japanese RT middleware OS: Different Versions/CPU word size Compile Software Running on Simulator Verify Test Results

31 Inheriting Humanoid Robots from RBR 2013 Remote Brain Inherit: Mother Sensor Suit/Flesh Evolve to More Complex body 3D Flow Onboard vision Lighter devices Common Integration Platform Inherit: Specialization Evolve with Open Community Musculoskeletal High Power Care Kitchen Assist Life Assist Jump Push Recovery Spine

32 How? Towards Whole-Body Humanoid Development Methodology Software Hardware Body Type 1983 Rope Handling 1 st ISRR 2013 Dancing Steps 1988 Dexterous Manipulation Environment Behaviors Task Integration Inheritance 2013 Musculoskeletal Humanoid

33 Inheriting Humanoid Systems from Remote-Brained Robotics Masayuki Inaba Professor, JSK Robotics Lab. Department of Mechano-Informatics/ Department of Creative Informatics University of Tokyo 1 Prof., 1 Associate Prof. 3 Assistant Profs., 35 students, 2013FY 33

34 Brain - Body Interface

35 HTML/Web based robot UI ROS message over http network enable us to create portable UI on the web browser Display all image view from PR2 MoveIt! (Planning Apps) Web Interface 3D Image Recognition result

36 HTML/Web based 3D UI Linux Industry Teaching Tablet Mac Browser-based UI 36

37 Humanoid Manipulation 2012

38 Whole-body version