4/1/2011. Ken Goldberg UC Berkeley. Robot

Transcription

1 The World of Robots history Ken Goldberg UC Berkeley 2 history Robot Karel Capek, R.U.R. (1923) 3 1

2 Two Classes of Robots Industrial robot : Reprogrammable, multi-function manipulator with 3 or more axes. Anthropomorphic Unstructured Flexible Non-Anthropomorphic Structured Efficient (Robot Institute of America) Tesla, 1898 Goertz, 54 Mosher, 64 Tomovic, 69 Salisbury, Bejczy, 85 Ballard, 86 Volz, 87 Sheridan, 92 Sato, 94 Presence Journal 92 O. Khatib, et al. 96 Tele-Robot s: cnc machines, industrial robots kinematics, dynamics 8 2

3 1980s robotic systems Adept's SCARA robots Cognex In-Sight Robot Barrett Technology Manipulator : phase change in adoption FIRST robotics 1000 surgical robots 10,000 US military robots 5,000,000 household vacuuming robots To transform our culture by creating a world where science and technology are celebrated and where young people dream of becoming science and technology leaders The FIRST vision 11 3

4 what s changed? Cost/Performance of Computing Quality of Sensors (Digital Cameras / LIDAR) Advances in Statistical Machine Learning Personal Robots Telerobotics Humanoids Open-Source kinect rgb-d camera robotics 1.0: robots replacing humans robotics 2.0: robots cooperating with humans 15 4

5 robotics 2.0: robots as Co-X Co-Workers: Manufacturing, Logistics, Healthcare Co-Inhabitants: Home Assistance, Rehabilitation, Driving Co-Protectors: Defense, Home Security, Monitoring robots as co-workers Kiva Systems 18 robots as co-workers cobot: colgate and peshkin robots as co-workers Aethon hospital robots

6 robots as co-inhabitants Roomba robots as co-inhabitants robot learning (Pieter Abbeel) Robo-cup robots as co-inhabitants Cyberdyne Hybrid Assistive Limb (HAL) Outline: 24 6

7 robots as co-inhabitants Google computer aided driving tele-actor Robots as co-inhabitants new telerobots 2010 Outline: anybot rovio 28 7

8 robots as co-inhabitants Hiroshi Ishiguro s Geminoid HI-1 robots as co-protectors Boston Dynamics Outline: robots as co-protectors xxxx robots as co-protectors learning from human demonstrations Outline: Outline:

9 robots as co-protectors Intuitive surgical Needle Minimization Minimize number of needles Less needles = less trauma Avoid sensitive organs Puncturing organs leads to negative side effects Speed Recovery 33 Possible Needles Minimal Subset Bevel-tip Circular paths Steerable Needles (awarded U.S. Patent in 2010) Motion planning for steerable needles target needle tip entry point External control by pushing and rotating base 35 External control by pushing and rotating base 36 9

10 learning from human experts for robot assisted surgery Motivation Repetitive Subtasks Retraction Suturing Robots can: Enhance surgeon performance Reduce tedium and medical errors Reduce time and costs Improve patient health Potential for Tele-Operation Da Vinci Objective Learning from Demonstrations Human approximates ideal trajectory But always imperfect, noisy, Multiple demonstrations Vary in different ways Implicitly encode ideal trajectory First step: infer ideal trajectory Human demonstrations 10

11 Dynamic Time Warping Demonstrations vary in quality: Matrices R specifying relative quality of demonstrations Demonstrations vary in length/duration Need to be time-aligned to match Simultaneous optimization of R, time warping, and ideal trajectory EM-algorithm Dynamic Time Warping Given: Inferring Ideal Trajectory Probabilistic dynamics model (x t+1, u N demonstrations as noisy observations (x t1, u t1,, x tn, u tn ) = C (x t, u t ) + z t, z t ~ N( Dynamics and observations of hidden `process Use Kalman Smoother to infer ideal trajectory Approach Results: 1. Learning from human demonstrations -> ideal trajectory 2. Robot learns to execute tasks with superhuman performance Iteratively optimize controller Increase speed Intended Human demonstrations Inferred task trajectory Execution 1x Execution 4x Execution 7x Execution 10x 11

12 u.s. robotics 2.0 initiative White House OSTP announcing R&D priorities for the President s FY2012 budget Report forthcoming: A Vision for Robotics 2.0 in America robotics 1.0: robots replacing humans robotics 2.0: robots cooperating with humans 45 The World of Robots Ken Goldberg UC Berkeley 12