Research Areas at the Intel Research Laboratory at Seattle

	Research in the lab is spread across four major projects: Personal Robotics Everyday Behavioral Monitoring Technology for Long-Term Care Trustworthy Wireless
	Personal Robotics The Personal Robotics Project is a collaborative effort of Intel Research Seattle, Intel Research Pittsburgh, and our colleagues at the University of Washington and Carnegie Mellon University. One motivation for the project is the business hypothesis that the robotics industry today is at a point analogous to the personal computing industry of the early 1980s. In the next decade, the number of personal robots deployed in unstructured environments, such as homes, could grow dramatically, if the right enabling technologies and "killer applications" are developed. The three-year goal of Personal Robotics is to build a robot that is able to locate, pick up, and retrieve particular objects in a home or office environment: "Robot, get me the {beer, medication, eyeglasses, pacifier…}" This problem -- and the field of robotics in general -- may be roughly divided into two parts: navigation and manipulation. The biggest obstacle to the three-year vision today is robotic sensing and perception, especially as it applies to manipulation. In recent years, the invention of the laser rangefinder has transformed robotic navigation. Last year's successful completion of the DARPA Grand Challenge was a milestone that clearly demonstrated how sophisticated robotic navigation has become. New sensing and perception techniques could bring similar benefits to the challenge of manipulation. At the Seattle lab, a team lead by Joshua Smith is building robotic graspers with a new sense: electric-field-based "pretouch," a sense that is shorter in range than vision but longer in range than touch. They have also developed a novel approach to the visual object instance recognition problem that benefits from recent scaling of storage and computing.
	Everyday Behavioral Monitoring The Everyday Behavioral Monitoring (EBM) Project seeks to enable wide-scale adoption of technologies that sense, model, and support the user in everyday activities. The project develops unobtrusive wearable devices and algorithms that can discern between activities such as running and climbing stairs, cooking dinner and working on a computer. The machines will even be able to perceive whether the user is engaged in a lively debate or telling a bedtime story. The challenge in this research arises from the inherent "noise" of our environment. The unpredictable variation of our everyday lives becomes an impediment to sensor data collection. By testing their devices in uncontrolled environments, project lead, Beverly Harrison and her colleagues, Sunny Consolvo and Tanzeem Choudhury, are working to build privacy-sensitive applications that can adapt to the broad spectrum of human activity. The first application to emerge from the project was the UbiFit Garden, which attempts to encourage regular physical activity through the use of on-body sensing, mobile displays, and mobile journaling. With UbiFit Garden, a garden blooms on the screen background of an individual's mobile phone as she performs physical activities throughout the week. At a glance, she can determine if she is having an active or inactive week (by the number of flowers), if she has incorporated variety into her routine (by the types of flowers), and if she has met her goals this week and in recent weeks (by the appearance of large and small butterflies). Physical activities are automatically sensed by a wearable device and manually added and edited through a journal on the mobile phone. Another arm of the team's research involves the creation of models that can decipher complex social interactions. Collecting speech in situ requires recording conversations in unconstrained and unpredictable settings, both public and private. But obtaining audio data from real-life situations is a tricky ethical and legal issue. To preserve privacy, EBM investigators have embedded privacy protection into the lowest levels of processing. They are developing techniques for interpreting the dynamics of a social encounter without requiring access to the raw data streams. Though the subject matter of the recorded speech is unintelligible to the human ear, it still provides enough data for the software to recognize different types of conversation, reason about the relationships between individuals, and infer the structure of social networks.
	Technology for Long-Term Care (TLC) Long-term care -- helping elders with tasks required for day-to-day functioning -- is a critical problem in many societies. In the US, the estimated cost of such care in the year 2000 was roughly $275 billion. The population of elders is expected to grow sharply in the next few decades, but the resources available for caring for them are not. Without a significant increase in productivity, many elders are therefore likely to be left without adequate care. Technology for Long-Term Care is a validation project aimed at showing that sensor-based monitoring of elder activity can potentially reduce the burden of care giving and increase the independence of elders. The project will deploy activity recognition technology developed by Intel in 20 elder residences (both homes and community care settings) in the Seattle area for three months beginning October 2007. Lightweight electronic displays will provide elders and their care givers with up-to-the-hour information on whether the elder performed any of six activities. The hope is that elders can use this information as a memory aid to keep track of activities performed (thus relying less on their care giver), and that care givers can keep an eye on the elder without physically being present in the elder's home or room. If successful, this formative study will be followed up by a much larger statistically valid study. TLC uses technologies developed by the earlier Human Activity Recognition (HAR) project at Intel Research Seattle. The key new ideas include a focus on sensing object use, ultra-dense sensing (based on RFID and compact wearable devices), statistical modeling with few examples (based on formalized common sense) and coordinated care display interfaces (which leverage inexpensive digital displays showing information appropriate to a person's network of care givers). These techniques have resulted in improvements in the number, detail, ease-of-modeling, and accuracy of activity recognition by a factor of ten to a hundred. TLC is a collaborative effort among Intel (both Research and the Digital Health business group), the Veterans Administration, and the University of Washington in close coordination with the Washington State Aging and Disabilities Services Administration (ADSA). Matthai Philipose heads this project.
	Trustworthy Wireless Wireless communications are becoming ubiquitous and personal as networked devices such as mobile phones, handheld PCs, cameras, music players, and health monitors are increasingly part of the everyday computing environment. In this setting, it is vital that people can use these devices with confidence and without leaving behind "digital footprints" of their identities and activities that can later compromise them. The Trustworthy Wireless Project, under the direction of Ben Greenstein, aims to eliminate the privacy concerns associated with wireless protocols. Wireless, mobile devices present a greater security risk than wired networks because wireless transmissions are broadcast to all nearby receivers, and because connectivity via mobile devices spreads user traffic over a wider range of parties. This risk is not addressed by the traditional approach of encrypting messages to ensure their confidentiality, nor by layering end-to-end encryption on applications. This is because the link and network layer designs contain low-level identifiers (e.g., MAC addresses) that often map directly to high-level identifiers (e.g., user names). The goal of the Trustworthy Wireless project is to discover new methods of privacy protection for the users of mobile devices from the physical layer up and across a range of wireless technologies. To accomplish this, the team measures existing systems such as RFID, 802.11, and WiMax to determine what private information is leaked in common usage and how this maps to user expectations. They are also designing revised protocols that offer stronger protections and expose the state of privacy to their users. Among other topics, Trustworthy Wireless explores: •tools that monitor transmissions and assess the state of privacy for users, including ways to check implementations online and with little hard-coded knowledge of applications. encrypted network protocols that scramble all low-level identifiers, including addresses and services names. These are normally be sent in the clear for the system to function, as they fall below or outside of traditional confidentiality and authentication mechanisms. ways to confine reception and connectivity to user-specified service areas at the physical layer, to enhance privacy, security from outside attack, and capacity.