Research

Sensing is becoming part of the spaces we live and work in, from wearables on the body to ambient sensors in homes and offices. Our goal is to make these systems understand people's activities and context in ways that are accurate, robust, and respectful of privacy. A recurring theme in our work is moving away from cameras and raw audio toward privacy-preserving modalities such as radar, lidar, thermal, IMU, and filtered audio, while still achieving rich recognition. For example, VAX uses existing video and audio models to bootstrap privacy-sensitive sensors without manual labeling, and OrganicHAR lets activity categories emerge from what the sensors themselves can reliably detect, using vision-language models only at key moments. Kirigami filters speech out of always-on audio sensing, TAO infers higher-level context from daily activity patterns, and IMUCoCo enables flexible on-body IMU placement for pose estimation and activity recognition. We are broadly interested in how sensing systems can be deployed in-situ, maintained over time, and adapted to new people and environments with minimal effort.

As connected devices and apps collect ever more data about people, we study how to make these technologies trustworthy, transparent, and controllable by the people they affect. Our work spans both systems and human factors. On the consumer side, we led the design of the IoT Security and Privacy Label, studied how label complexity, the U.S. Cyber Trust Mark, and QR codes affect consumer comprehension and purchase decisions, and helped inform national IoT labeling efforts; this line of work was recognized with the Norm Hardy Prize from the Foresight Institute. For shared and always-on sensing, we explore mechanisms for transparency and control in smart homes, and ThingPoll, which lets the people sharing a space negotiate privacy settings together. For developers, we build tools such as Matcha and Honeysuckle that help create accurate privacy nutrition labels and in-app privacy notices. We also study the security of the broader device ecosystem, such as granular data ownership for IoT devices.

Buildings, transportation, the electric grid, and computing itself account for a large share of global carbon emissions. As part of the NSF Expeditions in Computing project on Computational Decarbonization of Societal Infrastructure (CoDec), we develop computational methods to measure, reduce, and manage this carbon footprint (vision paper). Our work includes SpiderGen, which uses generative AI to produce procedures for carbon life cycle assessments, and studies of cost-effective incentives for siting solar deployments. We also look at the energy and carbon cost of AI: ECO-LLM orchestrates domain-specific language models across edge and cloud to balance accuracy against resource and energy use. This direction builds on the lab's long history of energy-efficient computing, from low-power PC sleep states to energy-aware mobile systems and smart buildings.

Buildings are where people spend most of their time and where much of the world's energy is consumed, which makes them a natural place to deploy sensing and computing at scale. For more than a decade, we have designed, built, and deployed the infrastructure for smart buildings. This includes BuildingDepot, an open-source building "operating system"; Brick, a metadata schema for portable building applications now supported by an industry consortium; and Mites, a general-purpose sensing infrastructure deployed across a campus building, whose paper received the Ubicomp/IMWUT Distinguished Paper Award. On top of this infrastructure, we explore how occupants and operators can interact with buildings. For example, BuildingChat enables natural-language interaction with building-scale sensing. Our open questions include how to make sensing substrates robust and easy to maintain, how to enable portable building apps, and how to balance utility with occupant privacy.