Progress and Opportunities in Robotics

Transcription

1 Gill Pratt Dr. Gill Pratt joined DARPA as a Program Manager in the Defense Sciences Office in January His primary interest is in the field of robotics. Specific areas include the development of declarative design methods that enhance the symbiosis between designer and design tool, hyper-rapid fabrication methods, interfaces that significantly enhance human/machine collaboration, mechanisms and control methods for enhanced mobility and manipulation, low impedance actuators, and improved platforms for post-secondary robotics education. He also has a strong interest in the application of neuroscience techniques to robot perception and control. Dr. Pratt holds a Doctor in Philosophy in electrical engineering and computer science from the Massachusetts Institute of Technology (MIT). His thesis is in the field of neurophysiology. He was an Associate Professor and Director of the Leg Lab at MIT. Subsequently, he became a Professor at Franklin W. Olin College, and before joining DARPA, was Associate Dean of Faculty Affairs and Research. Dr. Pratt holds several patents in series elastic actuation and adaptive control.

2 Progress and Opportunities in Robotics Dr. Gill Pratt Program Manager Prepared for UCSD May 8,

3 The Problem Distribution Statement 2

4 The IED Problem EOD missions in OIF and OEF, through Sep 2010 IEDs detected and defeated with UGVs 1 11,000 UGVs deployed 1 8,000 UGVs destroyed by IEDs EOD Technicians KIA 3 97 Source: IED Metrics for Iraq, Center for Strategic and International Studies, 22 July Source: IED Metrics for Afghanistan, Center for Strategic and International Studies, 21 July Unmanned Systems Integrated Roadmap, FY , pp As of Sep Minds of their own, Erik Schechter, C4ISR Journal, 09 Sep EOD Memorial Foundation, Through June 2011, 5/11/2015 3

5 Both Productivity and Protection Matter 100 Task Complexity Future Kiva Productivity Gain = #(People No Robot) #(People With Robot) 10 MDARS or #($ No Robot) #($ With Robot) 1 Talon Predator Protection Gain = P(Casualty No Robot) P(Casualty With Robot) 4

6 Progress 5

7 Autonomy 6

8 Mobility

9 Speed

10 Cost : ARM-H 9

11 Energy Efficiency of Vehicles + Robots Program Start: SR=5 Rough Terrain Robots 0.2 MMMMMMMMMMMMMMMMMMM IIIIIIIIII PPPPPPPPPP mm gg vv 100 X 2 20 X Program Goal: SR= Miles per Gallon 1 Speed (m/s) Speed (mph) Wheeled Tracked Legged Other Miles per Gallon for a Toyota Camry (1045 kg) running on gasoline, using an energy conversion efficiency of 25% 10

12 Energy Efficiency MIT Cheetah 11

13 Total Cost of Transport (P total /WV) Log Minimum cost of Transport, P/(WV) ASIMO (2) Human Running MIT Cheetah Robot (0.45) Cheetah Adopted from Tucker 1975 Log Mass, kg

14 The DARPA Robotics Challenge (DRC) 5/11/2015 Distribution Statement A (). 13

15 Why a Disaster Response Challenge? We are vulnerable to natural and man-made disasters HADR is 1 of the 10 primary missions of the US DoD Sustaining U.S. Global Leadership: Priorities for 21st Century Defense, The White House,+ SecDef, January 2012 Fukushima Daiichi, March 2011 close study of the disaster s first 24 hours, before the cascade of failures carried reactor 1 beyond any hope of salvation, reveals clear inflection points where minor differences would have prevented events from spiraling out of control. IEEE Spectrum, Nov 2011 p. 36 HADR is a universally understood and appreciated mission Enables participation of best and brightest performers, from anywhere in the world 14

16 Why a Challenge? Napoleon Bonaparte's Food Preservation Challenge Year: 1795 Prize: 12,000 Francs Winner: Jan, 1810 by Nicolas Appert (Jar) Runner Up: Aug, 1810 by Peter Durand (Can) Enabled canned food DARPA Grand + Urban Challenges Year: Prize (Winner) 2004: $1M (No Winner) 2005: $2M (Stanford) 2007: $2M, $1M, $0.5M (CMU, Stanford, VT) Enabled Google, Oshkosh Driverless Vehicles + several auto manufacturer efforts Public Internet Images Public Internet Image Public Internet Image 15

17 Human-Robot Partnership One Measure: AAAAAAAAAAAAAAAA KKKKKKKKKKKKKKKKKKKK CCCCCCCCCCCCCCCCCCCC oooo TTTTTTTT SSSSSSSSSSSSS EEEEEEEEEEEEEE oooo CCCCCCCCCCCCCCCCCCCCCCCCCC DRC Robot Degraded Network Emulator Operator 16

18 DRC Program Structure and Funding CY2012 CY2013 CY Phase 1 Phase 2 Track A System Track B Software Track C Software Track D System 7 Teams $1.8M each 11 Teams $375k each 115 Teams Self-funded 4 Teams Critical Design Review Virtual Robotics Challenge Self-funded from Industry, Academia 6 Teams $1.2M each 7 Teams $750k each + Atlas Robot 11 Teams $1.5M each 14 Teams China, Germany (2), Hong Kong, Italy, Japan (5), South Korea (2), US (2) 8 Months 6 Months 18 Months $2M 1 st Place $1M 2 nd Place $0.5M 3 rd Place October 2012 DRC Kickoff Note: Funding levels are approximate and vary by team. June 2013 December 2013 DRC Trials Homestead, FL June 5-7, 2015 DRC Finals Pomona, CA 17

19 Anticipated Robotic Challenge Trials Tasks Capability Exercised Autonomy - Perception Autonomy Decision-making Mounted Mobility Dismounted Mobility Dexterity Strength Endurance Sample Tasks 1. Drive utility vehicle (e.g. Gator, Ranger) X X X X 2. Travel dismounted 20 m through various terrains X X X 3. Remove debris blocking entryway X X X X X 4. Open door, enter building X X X X 5. Climb industrial ladder/stairs/walkway X X X 6. Break through wall X X X X X 7. Locate and close valve X X X X X X 8. Connect fire hose X X X X X 18

20 A (small number of) Highlights of Day 2 (Saturday) 19

21 Performer Results: Exceeded Expectations SCHAFT IHMC Robotics Tartan Rescue MIT RoboSimian TRACLabs WRECS TROOPER THOR ViGIR KAIST HKU DRC-Hubo Chiron NASA JSC Mojavaton Green: Team ranked in top 8 Maximum of 32 points possible 20

22 Front Page of the Asahi Shimbun, April 30, 2014 Google gobbles up Japanese robot technology Google's acquisition of the Japanese start-up Schaft Inc. last November was a major shock to the Ministry of Economy, Trade and Industry, which had been seeking out ways to cooperate with the U.S. government in robot technology development. U.S. President Barack Obama met with Schaft's young founders at the National Museum of Emerging Science and Innovation during his Japan visit last week. He kicked around a soccer ball with the humanoid Asimo robot and also observed a blue robot that was still in the developmental stage. Yuto Nakanishi and Junichi Urata explained to the president that their robot could perform functions in areas where humans cannot enter, such as parts of the crippled Fukushima No. 1 nuclear power plant. They both gave up their positions as assistant professors at the University of Tokyo to launch their company in Public Document 21

23 DARPA Robotics Challenge International Task Workshop, Washington, DC, September 29, 2014 John Holdren, Director of the White House Office of Science and Technology Policy (OSTP) Indian Treaty Room, Eisenhower Executive Office Building, the White House Dr. Chul Hue Park KIRIA Korean Lead Delegate Dr. Gill Pratt DARPA TTO Program Manager Dr. Brad Tousley DARPA TTO Office Director Mr. Noboru Aoki METI Japanese Lead Delegate Photographs by Mike Perschbacher 22

24 Bots for Finals (First 15) 23

25 Bots for Finals (Last 10) 24

26 Opportunities : A few of many 25

27 Near Future (but everybody s doing it): Autonomous Driving Public Internet Images 26

28 Aging Society 27

29 Manufacturing Nature pays by the pound, not for complexity Why can t we? Dollar a Pound, The Garment District, Cambridge, MA Photograph by Gill Pratt 28

30 Wireless Enabled Cloud Robotics World s data storage now measured in Zetabytes (10^21 Bytes) By Comparison Number of Synapses in Human Brain: ~ 10^14 About 10 billion images have been uploaded World s computing capacity approaching 1 Zeta OPS Google is one of world s largest consumers and manufacturers of computers Highest performance video games now do 80% of their computing on the cloud High speed wireless connection to the internet becoming ubiquitous Example Product: Google Chromecast ($35) Batteries have low energy density (approx. 1/10 fossil fuels) SWaP is at a premium in mobile devices A server room in Council Bluffs, Iowa. Photo: Google/Connie Zhou Hard part of robotics is between the ears (of the robot) Many problems get easier with lots of data + processing Example: Use of maps for autonomous driving Example: Visual object perception Big Idea : Put the robot brain on the cloud Side benefit all robots learn from each robot s experience We still needs to develop competency in: Unstructured, austere environments Intermittent Communications Better-than-human performance Low SWaP Limited a-priori knowledge Critical (human life) missions 29

31 Sparse Ultra-Low Power Computation 16 Million Neuron IBM True North SyNAPSE System Public Internet Image 30

32 31