The Embedded Reasoning Area in the Intelligent Systems Laboratory at PARC is a multidisciplinary group of people dedicated to research on improving the flexibility and intelligence of systems as computation becomes increasingly ubiquitous and embedded.
Overview
The adaptive appliance, the software-enabled car, the smart home, the reconfigurable factory - these are tomorrow's smart systems. With large numbers of embedded sensors, actuators, and processors, these distributed systems will be versatile, adaptive, and robust. They will be able to understand their environment and act intelligently, and often autonomously, to provide superior functionality. They will interact appropriately with each other and with their human operators. They will increasingly be built from smart components, configured just in time, and upgraded over time to suit changing requirements. All these properties will be achieved in large part through software-enabled control. Such software accurately determines a system’s state, intelligently reasons about its diverse capabilities, adaptively coordinates its components in the context of dynamic goals, and robustly controls its behavior by integrating long-range planning, optimal behavior selection, and reactive control.
Control means directing a system such that it achieves its objectives within its inherent constraints. Developing embedded control software is exceptionally hard due to complex component behaviors and interactions, incomplete and uncertain sensing of the environment, and not least the stringent real-time and resource constraints on computing and communication in the embedded system. Control software development is increasingly the leading cause for project time and cost overruns, and thus can be a major barrier to deploying advanced systems capabilities. We believe that the best hope for taming this complexity is software that builds on declarative models of the physical system, uses advanced reasoning in order to deduce and coordinate the appropriate actions, and is embedded in a formal architecture for integrating distributed intelligent components.
Research
Within the vision of embedded reasoning outlined above, the group currently focuses on the following research areas:
Constraint-based planning and scheduling: integrated planning and scheduling, real-time search, heuristic techniques, modeling languages
Distributed model-based control: hybrid control, control decomposition, distributed reasoning, interface contracts, autonomy
Model-based search and constrained optimization: adaptive search, problem-specific techniques, time-bounded techniques
Adaptive network architecture: large-scale architectures, physical control allocation, communication and negotiation protocols, performance analysis
This program collaborates with other Smart Matter Integrated Systems projects, including the Modular Robotics project. Also, in the context of DARPA's NEST program, we are developing techniques for distributed adaptive constrained optimization.
Vision of printer with many processors
Air jet paper mover
Localization in ad-hoc sensor networks
