Safe AI-enabled Autonomy (DARPA Assured Autonomy)
We are in the midst of a foundational shift. From self-driving cars and voice assistants to smart thermostats and recommendation engines, Artificial Intelligence (AI) and machine learning are becoming an integral part of our daily lives. The emergence of these technologies opens up countless opportunities to transform any industry and to revolutionize traditional ways of thinking, operating, and solving problems. But…how do you know if you can trust AI? Our group is focused on working with existing AI-enabled autonomous systems, making them safer, and providing rigorous guarantees of their safety. Our approach uses ideas from robust control and robust optimization (broadly defined) in order to make machine learning more robust to perturbed data or distributional shifts in the data. Quantifying the uncertainty of machine learning components is also critical as it allows systems designs that compensate for the uncertainty induce my machine learning components. Our efforts can be applied to more classical perception based tasks (classification) or in the safety and stability analysis of deep reinforcement learning of autonomous systems with machine learning components in the feedback loop.
Perception-based safe planning in unknown environments (AFOSR Assured Autonomy)
The majority of motion or mission planning algorithms for autonomous systems assume robots with known dynamics operating in known environments. As a result, these methods cannot be safely applied to scenarios where the environment is initially unknown but but is continuously perceived and mapped using recent advances in machine learning. To address these challenges, we are pursuing developing perception-based complex mission planning algorithm for multi-robot systems with known dynamics that operate in uncertain environments. Specifically, the uncertain environment is modeled using probabilistic semantic maps and/or occupancy grid maps while mission. Safety specifications are rigorously expressed in suitable temporal logic formalisms. Our approaches generate reactive control policies that adapt to the continuously learned map of the environment that is updated machine learning-based perception systems. The proposed method scales well with the number of robots and the size of workspace. We are also developing learning-based approaches for temporal logic mission planning for robots with unknown dynamics operating in uncertain environments.
Estimation and Control over 5G networks (NSF CPS), Intel Science and Technology Center
The internet-of-things (IoT) revolution is bringing millions of physical devices online (e.g. cars, UAVs, homes, medical devices), enabling them to connect to each other in real-time, as well as to cloud services. Beyond 5G wireless communication will be critical in providing IoT connectivity. Our project focuses on low-latency and ultra-reliable communications and networking that is critical for latency-sensitive, closed-loop control applications, like vehicle to vehicle communications, collaborative swam planning, and Industry 4.0 applications. In such latency sensitive applications, we do not know what is possible and what are the fundamental limits for control system design over low-latency, high-reliability communications. In this project, we will be rethinking the scientific foundations for ultra-reliable, low-latency wireless communications for latency sensitive control applications. We propose to achieve our scientific agenda by addressing three intellectual challenges: 1) Low-latency channel coding, where the goal is to focus on short packet codes for control loops 2) Control over low latency-aware communication channels, where the goal is to understand the what is the optimal tradeoff of latency to reliability for control loops and 3) Learning for Large Scale Wireless Control Networks, where machine learning will perform resource allocation for large numbers of control loops with competing latency/reliability requirements We intend to evaluate the proposed research agenda by leveraging our existing Intel Science and Technology Center (ISTC) on Wireless Autonomous Systems and demonstrate our ideas in future wireless protocols (IEEE 802.11ax) and experimentally demonstrate it in high-speed V2V and fast formation control with aerial swarms.
Resilience of Critical Networked Infrastructure (The Rockefeller Foundation)
Multi-agent networked systems are used to model a wide variety of systems, from robotic networks to power systems, to biological networks. The emergent dynamics of a network of dynamic agents can be strongly affected by the presence of network failures. The main of our research in this direction is to find computationally efficient methods to study the effects of link or node failures on networked infrastructure. We have provided several methods for detection and isolation of failures in a network of dynamic agents, based on the presence of discontinuities in the derivatives of the output responses of a subset of nodes.
Security and Privacy-Aware Cyber-Physical Systems (NSF, Intel)
The project aims to achieve a comprehensive understanding of CPS-specific security and privacy challenges. This understanding will enable us to (1) develop techniques to prevent security attacks to CPS and to detect and recover from malicious attacks to CPS; (2) develop techniques for security-aware control design by develop attack resilient state estimator; (3) ensure privacy of data collected and used by CPS, and (4) establish an evidence-based framework for CPS security and privacy assurance, taking into account the operating context of the system and human factors.
The TerraSwarm Research Center (DARPA FCRP)
Over the past decade there has been increasing interest in the use of swarms of mobile sensors to help solve societal-scale problems. Sensor swarms, which can be wirelessly interconnected and deposit vast quantities of data in centralized repositories, offer an unprecedented ability to monitor and act on a range of evolving physical quantities. TerraSwarm applications are characterized by their ability to dynamically recruit re- sources such as sensors and data from the cloud, aggregate and use that information to make or aid decisions, and then dynamically recruit actuation resources – mediating their response by policy, security, and privacy concerns. The TerraSwarm Research Center (TSRC) aims to enable the simple, reliable, and secure deployment of advanced distributed sense-control-actuate applications on shared, massively distributed, heterogeneous, and mostly uncoordinated swarm platforms through an open and universal systems architecture.
Synthesis of Platform-aware Attack-Resilient Control Systems (DARPA HACMS)
Embedded systems form a ubiquitous, networked, computing substrate that underlies much of modern technological society. Such systems range from large supervisory control and data acquisition (SCADA) systems that manage physical infrastructure to medical devices such as pacemakers and insulin pumps, to computer peripherals such as printers and routers, to communication devices such as cell phones and radios, to vehicles such as airplanes and satellites. Such devices have been networked for a variety of reasons, including the ability to conveniently access diagnostic information, perform software updates, provide innovative features, lower costs, and improve ease of use. Researchers and hackers have shown that these kinds of networked embedded systems are vulnerable to remote security attacks, and such attacks can cause physical damage while hiding the effects from monitors. The goal of the HACMS program is to create technology for the construction of high-assurance cyber-physical control systems, where high assurance is defined to mean functionally correct and satisfying appropriate safety and security properties.
Assuring the Safety, Security and Reliability of Medical Device Cyber Physical Systems (NSF CPS)
The objective of this research is to establish a new development paradigm that enables the effective design, implementation, and certification of medical device cyber-physical systems. The approach is to pursue the following research directions: 1) to support medical device interconnectivity and interoperability with network-enabled control; 2) to apply coordination between medical devices to support emerging clinical scenarios; 3) to close the loop and enable feedback about the condition of the patient to the devices delivering therapy; and 4) to assure safety and effectiveness of interoperating medical devices. Novel design methods and certification techniques will significantly improve patient safety. The introduction of closed-loop scenarios into clinical practice will reduce the burden that caregivers are currently facing and will have the potential of reducing the overall costs of health care.
Micro Autonomous Systems & Technology (ARL CTA MAST)
The Micro Autonomous Systems Technologies (MAST) consists of four research centers focusing on microsystem mechanics, microelectronics, autonomous operations, and systems integration. The objective of the consortium is to develop autonomous, collaborative ensembles of agile, mobile microsystems to enhance tactical situational awareness in urban and complex terrain for small unit operations. The University of Pennsylvania is leading in the autonomous operation research center. Our emphasis is on processing for autonomous operations, which will provide the fundamental underpinnings for autonomous operation of distributed, mobile, multi-modal sensing micro-systems.
Expeditions in Computer Augmented Program Engineering (NSF Expeditions)
The goal of ExCAPE is to transform the way programmers develop software by advancing the theory and practice of software synthesis. In the proposed paradigm, a programmer can express insights through a variety of forms such as incomplete programs, example behaviors, and high-level requirements, and the synthesis tool generates the implementation relying on powerful analysis algorithms and programmer collaboration. The ExCAPE plan is to produce a range of design tools; that let end users program robots by demonstrating example behaviors, and that provide smart assistance for expert programmers to meet challenges in multicore programming.
Greater Philadelphia Innovation Cluster (GPIC) for Energy Efficient Buildings (DoE HUB)
The Greater Philadelphia Innovation Cluster (GPIC) for Energy Efficient Buildings received $129 million from the Federal Government’s Energy Regional Innovation Cluster (E-RIC) Initiative. The award included $122 million from the U.S. DOE to create the GPIC/HUB to develop innovative energy efficient building technologies, designs and systems. The GPIC/HUB goal is to transform the commercial building retrofit and new construction processes into a systems-delivery industry, and demonstrate building operational energy savings of 50% by 2013-2015 in a scalable, repeatable and cost effective manner across a broad building stock, while preserving workplace quality.
Heterogeneous Unmanned Networked Teams (ONR HUNT)
The grand challenge for this multi-university project is to push the state-of-the-art in complex, time-critical mission planning and execution for large numbers of heterogeneous vehicles collaborating with humans. Sophisticated cooperation among intelligent biological organisms, including humans, will offer critical insight and solution templates for many hard engineering problems. To meet this challenge, we have a assembled an interdisciplinary team of leading researchers who have pioneered work in artificial intelligence, vehicle control and robotics, cognitive psychology and human factors, biology, and political economics. This multi-university project is led by the University of Pennsylvania and will be performed in collaboration with the Georgia Institute of Technology, the University of California at Berkeley, medicalnd Arizona State University.
Quantitative Analysis and Design of Control Networks (NSF CPS)
Control networks are wireless substrates for industrial automation control, such as the WirelessHART and Honeywell’s OneWireless, and have fundamental differences over their sensor network counterparts as they also include actuation and the physical dynamics. The approach of the project is based on using time-triggered communication and computation as a unifying abstraction for understanding control networks. Time-triggered architectures enable the natural integration of communication, computation, and physical aspects of control networks as switched control systems. The time-triggered abstraction will serve for addressing the following interrelated themes: Optimal Schedules via Quantitative Automata, Quantitative Analysis and Design of Control Networks: Wireless Protocols for Optimal Control: Quantitative Trust Management for Control Networks. Our results will be integrated into control networks that are compatible with both WirelessHART and OneWireless specifications.
Situation undersanding bot through language and environment (ARO MURI)
This multi-university project brings together experts in linguistics, computational linguistics, artificial intelligence, machine learning, and robotics with the goal of developing a language for communication between humans and robots, where robots are responding to a changing environment and reporting on the relevant aspects of those changes. We will develop a broad range of concepts, which will ultimately enable designers to develop powerful communication methods between robots and humans, enabling humans to communicate goals and intentions as well as direct commands to robots in a natural, effective way. This multi-university project is led by the University of Pennsylvania and will be performed in collaboration with the University of Massachusstts at Amherst and the University of Massachusets at Lowell.
Robust testing by testing robustness of embedded systems (NSF EHS)
In recent years, the idea of the model-based design paradigm is to develop design models and subject them to early analysis, testing, and validation prior to their implementation. Simulation-based testing ensures that a finite number of user-defined system trajectories meet the desired specification. Even though computationally inexpensive simulation is ubiquitous in system design, it suffers from incompleteness, as it is impossible or impractical to test all system trajectories. On the other hand, verification methods enjoy completeness by showing that all system trajectories satisfy the desired property. This project brings together leading experts in embedded control, hybrid systems, and software monitoring and testing to develop the foundations of a modern framework for testing the robustness of embedded hybrid systems. The central idea that this project is centered around is the notion of a robust test, where the robustness of nominal test can be computed and used to infer that a tube of trajectories around the nominal test will yield the same qualitative behavior.
Scalable swarms of autonomous robots and mobile sensors (ARO MURI)
This multi-university project brings together experts in artificial intelligence, control theory, robotics, systems engineering and biology with the goal of understanding swarming behaviors in nature and applications of biologically-inspired models of swarm behaviors to large networked groups of autonomous vehicles. Our main goal is to develop a framework and methodology for the analysis of swarming behavior in biology and the synthesis of bio-inspired swarming behavior for engineered systems. This multi-university project is led by the University of Pennsylvania and will be performed in collaboration with the Massachusets Institute of Technology, the University of California at Berkeley, the University of California at Santa Barbara, and Yale University.
Synthesis of embedded software from hybrid models (NSF EHS)
Despite the proliferation of embedded devices in almost every engineered product, development of embedded software remains a low level, time consuming and error prone process. This is due to the fact that modern programming languages abstract away from time and platform constraints, while correctness of embedded software relies crucially on hard deadlines. This NSF-funded research aims at developing novel model-based design and implementation methodology for synthesizing reliable embedded software. Hybrid systems models, which allow mixing state-machine based discrete control with differential equation based continuous dynamics, are used for design and analysis. The research explores ways of mapping such models to code guided by correctness, modularity and portability issues. Technical challenges include bridging the gap between the platform-independent and timed semantics of the hybrid models and the executable software generated from it. This includes integrating generation of control tasks with scheduling to ensure optimal performance.
Sensor topologies for minimal planning (DARPA SToMP)
The Sensor Topology for Minimal Planning (SToMP) program leverages high-dimensional mathematical insights to create new capabilities that capitalize on emerging opportunities where sensors are miniaturized, pervasive, and coordinated. By developing new mathematical methods and techniques, the program seeks to revolutionize how networked sensors and autonomous agents are analyzed, distributed, and controlled. Of central importance will be a systematic determination and exploitation of minimization of total sensor complexity. More precisely, given a mission having a variety of costs to be optimized (e.g., total number of sensors, network bandwidth, or power consumption) the program seeks to determine solutions that require the least resources with respect to such metrics. Thus, the program aims to derive optimal solutions in that it focuses on implementations that are minimal in terms of reducing the total sensor complexity to the minimal configuration required.
Algorithmic synthesis of embedded controllers (NSF EHS)
Embedded systems require very novel, very challenging specifications that have to deal with synchronization, sequencing, and temporal ordering of different tasks. Mathematically formulating such desired specifications cannot be achieved using traditional mathematical formulations in control theory. On the other hand, computer aided verification has popularized the use of several temporal logics to describe complex specifications. However, the emphasis has been on verification of these properties for purely discrete systems, and not on synthesis for systems with a continuous component. In this research, a novel approach for automatically synthesizing hybrid controllers is pursued in order to satisfy specifications that are expressed in temporal logics. The proposed framework will provide algorithms and tools for the computation of discrete controllers, which by refinement will lead to embedded, hybrid controllers for the original system while providing performance and correctness guarantees.
Multi-robot emergency response (NSF ITR)
This project, in collaboration with the University of Minnesota and the California Institute of Technology, addresses research issues key to an important application of robot teams and information technology (emergency response in hazardous environments for various tasks). The research focuses on the development of methods for team coordination and dynamic distribution of tasks to robots. The project integrates the algorithms with first responder teams, emphasizing realistic scenarios.
Formal analysis and design of hybrid systems (NSF ITR)
High-level design of embedded software requires modeling concepts such as hierarchy, modularity, reuse, compositionality, and object-orientation. In this project we will develop a theory of hierarchical hybrid systems with an accompanying a compositional calculus of refinement. This will be the basis for behavioral interfaces and descriptions of components at different levels of abstractions. For rigorously specifying and evaluating design alternatives and correctness requirements, automated techniques such as model checking are very effective. To apply these techniques for formal analysis of hybrid systems, this research is developing automated schemes for constructing abstractions of hybrid models. The technical directions being pursued include model checking algorithms that exploit hierarchy, algorithms for extracting finite-state approximations using predicate abstraction, counter-example guided refinement of abstractions, property-preserving bisimulation-based reductions of continuous differential equations, and assume-guarantee reasoning. The results of this research are being integrated in software tools for modeling and analysis of hybrid systems. The benefits of the techniques for developing embedded systems with higher assurance for safety and reliability are evaluated in an experimental testbed of multiple, autonomous, mobile robots.
Adaptive coordinated control of intelligent multi-agent teams (ARO MURI)
This multi-university project involves the University of Pennsylvania, the University of California at Berkeley, and Carnegie Mellon University. It focuses on the design and evaluation of the adaptive hierarchical control of mixed autonomous and human operated semi-autonomous teams that deliver high levels of mission reliability despite uncertainty arising from rapidly evolving environments and malicious interference from an intelligent adversary. The design of architectures combining both hierarchical and heterarchical elements, the analytical foundations of interacting hybrid systems, the design of controllers for such systems that are robust against uncertainty, the management of rich sensory information from networked sensors among distributed and mobile teams; and the incorporation of human intervention in a mixed-initiative system are all key areas of our work. Additionally, the novelty of our approach is to explicitly take into account the need to adaptively replan missions to take into account environmental uncertainties and the deliberate malicious actions of a determined adversary. Equipment for this project is supported by an ARO DURIP grant.
Hierarchical abstractions of hybrid systems (NSF CAREER, PECASE)
Hybrid systems provide a theoretical foundation for the modeling, analysis, and design of embedded systems. Hybrid systems naturally combine discrete-event and continuous-time systems in a manner that can capture software logic, physical dynamics, and communication protocols, in a unified modeling framework. The wide applicability of hybrid systems has inspired a great deal of research from both control theory and theoretical computer science. Despite the great success of hybrid systems as a model, the applicability of state-of-the-art analysis and design techniques for hybrid systems has been limited to examples of small size due to complexity. Tthe research and educational agenda of the proposed research focuses on developing the theoretical foundations for the hierarchical decomposition of hybrid systems at various levels of abstraction. The long term goal of the research agenda will address the fundamental problem of given a class of hybrid models, and a class of properties that must be preserved, extract modeling abstractions that preserve the properties of interest. Achieving this goal will consist of first developing robust notions of bi-simulation for purely continuous systems, and then unifying the continuous and discrete notions in a manner that is consistent with the dynamics of hybrid systems.
Model based integration of embedded software (DARPA MoBIES)
The objective of this project is to develop a methodology and toolkit for design of embedded software for multi-agent communicating hybrid systems. The methodology will cover various design stages, including modeling, simulation, analysis, implementation, and monitoring. The methodology will be based on formal modular and hierarchical semantics of multi-agent hybrid systems. The methodology will improve the process for hybrid systems design by decreasing development costs through high-level modeling, by improving reusability by means of modular designs, and by developing more reliable designs with the help of analysis and runtime valiadation.