Exploratory Research at Intel

Transcription

1 Exploratory Research at Intel Ryan X. Wu, Ph.D. 吴欣然博士 Director, China Intel Research Corporate Technology Group 2008 Intel Corporation

2 Agenda Introducing Intel Research Essential Computing -Simplifying and enriching all aspects of work and daily life Everyday Sensing and Perception -Driving exciting advances in awareness which enables new capabilities 2

3 Intel Research Mission Drive off-roadmap, high-impact exploratory research vital to Intel Exploratory Research World class technical expertise Multi- disciplinary teams Open Collaboration, university ties 3

4 Engaging the Research Community Worldwide Close University Collaboration (Labs) Deep University Relationships (Focus Schools) Joint External Research Projects (Research Counci Israel Seattle PaPR Berkeley Santa Clara IBL Pittsburgh All over North America Europe, Asia, 4

5 Research through University Collaboration Intel Research Lablets on Campus Fund university research through CRC Intel Fellowship Program Focus School Program Large-scale Collaboration Program - UPCRC (Intel/Microsoft + UIUC/UCB) - Intel/HP/Yahoo Could Computing Testbed 5

6 Agenda Introducing Intel Research Essential Computing -Simplifying and enriching all aspects of work and daily life Everyday Sensing and Perception -Driving exciting advances in awareness which enables new capabilities 6

7 Essential Computing Things Essence you would of depend your life on Essential Simplifying and enriching all aspects of work and daily life 7

8 Personal Awareness Empower me to achieve the goals I value most Richly Communicative Easily form and enrich relationships Essential Computing Physicality Actuating everyday objects Concealing Complexity Technology that just works Data Rich Unleashing Internet-scale data Biosensors Coupling Bio and Info Systems 8

9 Personal Robotics Physicality Physicality Problem Statement Hypothesis: A Personal Robotics industry comparable to the Personal Computing industry is ready to emerge; the key problem for Personal Robots is taming uncertainty Approach: Develop new sensing, machine learning, and planning techniques to enable manipulation of real-world objects in natural settings such as a home Research Activities Building Physical Grep: Robot, get me the {soda, medication, eyeglasses, pacifier} - Sensing: Electric Field pretouch for hands - Perception: Novel data-driven object instance recognition technique - Planning: openrave: open-source test bed for planning; high dimensional motion planning - Manipulation: Learning to grasp from human demonstration; Grasp planning in cluttered scenes Electric Field Sensing Pretouch Boss: winner of 2008 Urban Challenge Recognizing objects in clutter Grasp planning in clutter executed in openrave The Robotic BarKeep autonomously perceives, plans, and manipulates in real-time 9

10 Personal Robotics Demo at IDF Aug. Joshua Smith, a principal engineer from Intel Research Seattle, passed an apple to a robotic hand that used electric field pretouch, a sense that fish use to detect their environment. Found it, the robot said. After grasping the apple on its own, Got it, the robot added. 10

11 Wireless Power Problem Statement Hypothesis: Wireless Power can be a compelling Intel platform ingredient Approach: Two approaches: low power (mw peak, uw average) far-field UHF- RFID-compatible sub-circuits for zeropower configuration; high-power (1W) near-field strongly coupled HF magnetic resonances Research Activities Low Power (WISP): Built world s first (1) UHFpowered accelerometer (e.g. wirelessly powered Wii game controller), (2) strain gage, (3) RFID tag with strong crypto. Demonstrated read, write, & erase of NOR Flash using in-package antenna via Near Field UHF coupling High Power: Modeling & building HF magnetically coupled hi Q resonators. Goal: demonstrate wireless laptop recharge Physicality Wirelessly powered accelerometer input device Voltage as a function of distance for long range UHF power Simulation & prototype high power magnetic resonance wireless power apparatus 11

12 Wireless Power at IDF this Aug Alanson Sample, an intern at Intel Research Seattle showed off this prototype, which uses resonance to transmit electricity wirelessly. 12

13 Log-Based Architectures Concealing Complexity Problem Statement Hypothesis: Architectural support can significantly improve the performance of online software correctness-checking tools (lifeguards). Approach: As application runs, hardware captures execution log; log is transported via processor cache and consumed by lifeguard running on separate core. Research Activities Design of architecture support for logging. Design/evaluation of log compression. Development of LBA simulator. Port of lifeguards to LBA. Linux system software implementation. Design of acceleration hardware. Detailed performance evaluation. Design for lifeguard parallelization. Design for monitoring parallel applications (on-going). 13

14 Programmable Matter Demo at IDF Aug. Jason Campbell, a senior staff research scientist at Intel Research Pittsburgh, showed models of what might be possible with programmable matter, which could change shape and color and move on its own. 14

15 Big Data Computing: Thriving in a world awash in data Our world is experiencing a flood of data from many diverse sources This vast collection of big data is an opportunity for disruptive change, if we can understand and use it. 15

16 Example Big Data Applications Digital photo scene completion Hays & Efros (CMU) Automatic gesture recognition in videos Ke (CMU/Intel), Sukthankar (Intel), Hebert (CMU) Seismic ground modeling Schlosser, Ryan, and O Hallaron (Intel) Interactive Search Assisted Decision Support (ISADS) Chen, Sukthankar, Mummert (Intel), with physicians at UPitt Medical Center and UPMC 16

17 Cloud Computing Test bed Intel/Yahoo/HP a globally distributed, Internet-scale testing environment designed to encourage research on the software, data center management and hardware issues associated with cloud computing at a larger scale than ever before. Six Centers of excellence cores per site. The test bed locations are expected to be fully operational and made accessible to researchers worldwide through a selection process later this year. 17

18 Tashi: Cloud Computing on Big Data IR Pittsburgh is developing Tashi, an open source cluster mgmt system for cloud computing on Big Data. Apache Software Foundation incubator with a development community from industry and academia Joint with Yahoo! and CMU Capability On-demand provisioning of the compute and storage resources Open interfaces to storage and compute management Compute and storage management working in concert Flexible compute and storage models Approach Build on existing cluster mgmt work such as Usher, COD, and EC2/S3. Develop the new capabilities that will be required to support big data cloud computing. Target audience Production system for groups with medium to large scale clusters Extensible research platform for distributed systems researchers 18

19 Agenda Introducing Intel Research Essential Computing -Simplifying and enriching all aspects of work and daily life Everyday Sensing and Perception -Driving exciting advances in awareness which enables new capabilities 19

20 Weiser s Vision of Ubiquitous Computing The most profound technologies are those that disappear, that weave themselves into the fabric of everyday life until they are indistinguishable in it. - paraphrase from Mark Weiser, The Computer for the 21 st Century, 1991, Scientific American 20

21 Everyday Sensing and Perception (ESP) Drive fundamental research advances that enable computing systems to become aware in everyday activities and environments Intel Research Imaging, Vision Machine learning Universities Data intensive Human computer interface Requirements for broad use: Accuracy, Coverage (variety, environment), Energy-Efficient, Privacy Preserving Achieving 90% accuracy for 90% of your day would move inference from hand-crafted scenarios into broad use (mainstream applications) Sensing Novel apps Research Laboratories ESP: Making computing systems aware of their users and context in everyday activities and environments 21

22 What People do Everyday Laugh Learn Touch Move 22

23 ESP Challenge: Efficient Sensing Accelerator Gyrometer Light Camera Humidity Sensing RFID All must sensors be = 3W managed Bluetooth Handheld < 1.5 hours for energy efficiency Barometer GPS WiFi Laser 23

24 ESP Challenge: Deriving High Level Understanding Actions Suggest, Avoid Help, Coach Move Toward Focus Sensing High Level Semantics Clean, Segmented Data Walk, Leave, Clean Happy, Angry Dangerous, Friendly He Said I Don t Like You Raw Sensor Data 24

25 ESP Challenge: Real-time Performance Actions High Level Semantics Suggest, Avoid Help, Coach Move Toward Focus Sensing Clean, Segmented Data Walk, Leave, Clean Happy, Angry Dangerous, Friendly He Said I Don t Like You Real-time video event detection requires 4 Teraflops at 10kW today. In the future, <1W on handheld? Raw Sensor Data 25

26 More Information Research at Intel - Exploratory Research at Intel - Intel Research Network of Laboratories - UPCRC

27 27