Daniel E. Koditschek - US grants

This planning grant is to redesign the Electrical and Systems Engineering (ESE) instructional programs at the University of Pennsylvania. Conceived as a "tree of learning," the new program unifies the diverse degree pathways through a common, stable, narrow "trunk" of newly conceived, required courses in the freshman and sophomore years - the focus of proposed effort. "Branching" out into specialties inherited from the various legacy programs in junior and senior years (and beyond), the new curriculum adds continuing attention to the "roots" in the form of a service-based learning component incorporated into key required courses throughout the four years. The dual concerns of this re-invention are to articulate the new intellectual foundations of a 21st Century undergraduate ESE degree and, simultaneously, to coordinate throughout the entire program the careful attention to human resource development (recruitment at the "roots" and retention through the "trunk" of the tree) now widely acknowledged as vital to the national interest.

The new ESE program aims to produce students who develop beyond technical mastery to become innovators, adept enough to lead the global market of ideas and products and adaptable to changing technology throughout their careers because of their comfort with self-directed learning. Adopting a research platform - the RHex robot - as the unifying lab focus in the early "trunk" semesters of the major will increase the rate of undergraduate participation in original research. The diversity of learning styles supported by such project-based, contextualized courses promises an increased retention rate, particularly among women and underrepresented minorities. Group work (project proposals, progress presentations, and so on) addresses the widely perceived need to bolster communication and writing skills in engineering training.

Intellectual Merit
This project mixes results and methods from cognitive psychology, computational vision and learning, neuromechanical systems biology, and robotics to develop a computer assisted environment for studying animal sensorimotor strategies, discovering how they undergird animal cognitive capabilities, and using those insights to inspire new algorithms for robot navigation, localization and situational awareness. We observe live, intact, highly mobile terrestrial invertebrate predators such as ghost crabs, desert scorpions and tiger beetles in carefully constructed habitats that challenge their ability to negotiate terrain and navigate space. We automate the collection, annotation and mathematical model extraction of their behavior from massive, parallel real-time recordings of visual, muscle, neural, and biomechanical recordings. We mine these data sets to develop intuitive hypotheses as well as formal mathematical representations of the basis on which these animals organize their own sensorimotor data streams to compile novel behaviors from previously consolidated constituents in a process of autonomous mental development. We add numerous existing sensor suites to highly agile existing robot bodies and instantiate algorithmically the hypothesized animal models to develop supporting or refuting evidence that challenges and refines them.
Broader Impacts
Scientifically, the new computational tools and ideas we identify in the interrelations we set up promise a bridge between whole areas of disciplines that have long been divided by spatiotemporal scale and the concomitant gap in analytical tradition, terminology and methods. For example, the study of these complex competencies in simpler species offers a new glimpse at the building blocks of cognition in species more closely related to humans. From the perspective of technological invention, algorithms pioneered in this research could lend an animal-like quality to a machine?s proximal tenacity in engaging its environment and even its overall situational awareness within unstructured worlds. For example, the team is inspired to imagine what it might be like to have a search and rescue robot with the (taskable) capabilities of a ghost crab. From the perspective of training and education, the automated database collection and management tools developed in this project bring to a mass audience the conceptual and computational building blocks that have heretofore been the exclusive province of a small group of experts. For example, a universally accessible (?cloud-based?) tool for unifying the design, parsing, display, and cross comparison of robots and animals searchable at will from the most intimate to the broadest scale of design and operation would have a profound impact on the ability of teachers at many different levels to motivate the fascination and unity of both synthetic and biological science.

This INSPIRE project is supported by the Directorate of Computer and Information Science and Engineering, the Division of Information and Intelligent Systems, the Directorate of Geosciences, Division of Earth Sciences, and the Office of International and Integrative Activities. Sand and dust storms menace the world. They impact large human populations on nearly every continent, damage habitation, disrupt transportation, threaten agriculture, biodiversity, human health and life, and degrade the environment (desertification). A team of scientists and engineers is developing an autonomous legged robot research assistant, designed to operate within harsh desert environments for purposes of gathering heretofore unavailable measurements of wind and sand movement under conditions far too uncomfortable and dangerous for human presence. These data may offer new insights into dust production and its global effects that could significantly impact predictions of environmental degradation by climate change in drylands. At the same time, meeting the formidable mobility and perceptual capabilities arising from scientists' requirements will advance the foundations and practice of robotics. Ultimately, advances in control of dust emissions from soils made possible by this novel collaboration could prolong the sustainability of the agroecosystem and result in improved air quality of downwind population centers.

Dust emission has traditionally been assumed to release sediment previously deposited in soil, but there is growing evidence that sand abrasion may actually produce significant quantities of new dust. Establishing sand seas as dust factories, rather than simply dust reservoirs represents a qualitatively new result that would cascade through aeolian and climate science. But the severe events of interest are hidden from existing conventional instruments and they pose presently insurmountable challenges to flying, wheeled or even tracked vehicles. Aeolian science needs legged machines to negotiate the steep, shifting slopes, and broken ground under the environmental conditions of interest. Steady running over simple terrain is a largely solved problem in robotics, but transitional maneuvers and agile negotiation of geometrically complex, unstable, fragile terrain characteristic of desert substrates pose a daunting next challenge for legged locomotion, requiring new approaches to turning, gait control, and perception.

This funding establishes a new Research Experience for Teachers (RET) Site at the University of Pennsylvania. The primary objective of this RET site is to involve middle school teachers in the School District of Philadelphia in summer research experiences that emphasize robotics. The teachers will spend six weeks in the summer participating in research experiences and developing classroom modules and materials which will be implemented in their classrooms during the academic year. The project is led by the General Robotics, Automation, Sensing, and Perception (GRASP) Laboratory at the University of Pennsylvania. The GRASP Lab is known not only as a leader in robotics research but also for its significant outreach and impact on K-12 education. The RET Site project involves a partnership between the university, the school district, the Mayor's Office of Education, and industrial associates. The teachers will continue to receive classroom support for related activities in their classrooms during the academic year including establishing FIRST LEGO League clubs in the schools. This RET Site project will develop a community of teachers who are passionate about robotics and who can translate this excitement to their students through engaging, high-quality inquiry learning experiences.

The project is anchored by a research area this is compelling and exciting for teachers and middle school students and by a faculty team that has demonstrated expertise in both research and education. The goals of the project are to: provide a quality research experience for middle school teachers in the field of robotics and automation; increase understanding of science and robotics; increase teachers' abilities to apply scientific and engineering methods to problems; increase the opportunities in underserved Philadelphia schools to prepare students for the engineering and scientific workforce; and develop teachers into STEM leaders. The site features projects that are teacher accessible as well as connected to current research and practice. The RET site includes sound evaluation and dissemination plans that will provide models for integrating robotics concepts into the middle school curricula. Teachers will develop and hone their research, communication, and presentation skills, all of which are essential to their professional growth and success. The project will build the technical capacity of teachers so they are capable of developing and implementing new, exciting robotics learning activities at their schools. The project team will disseminate the project materials through the TeachEngineering digital library and a GRASP's engagement web site that will be shared with a wider audience of teachers at annual workshops and conferences.

Sand and dust storms are a growing worldwide menace, soil instability and erosion threatens agriculture, and fine sediment compromises ecosystem health of rivers and oceans, all impacting large human populations on nearly every continent. The cumulative effects of these disturbed environments also threaten human well-being through damage of habitation and disruption of transportation. An interdisciplinary collaboration of geoscience, cognitive science, and robotics researchers aims to accelerate and deepen the collection of data about the fluid and materials properties associated with such unstable soils by endowing legged robots with the instrumentation and scientific agenda of agile, novice field assistants. These new robots are capable of general field mobility and are being programmed to think like assistant field geologists in order to develop research strategies in rugged natural environments where measurements are lacking. Overcoming the specific locomotion challenges presented by these environments and developing algorithms and software sufficient to interpret and act on human research needs will greatly advance the field of robotics. The resulting new information about wind, water and materials processes will have bearing on the management of infrastructure and agriculture, and also the response of landscapes to environmental changes.

The project focus is to use the geosciences field research setting to test a chain of hypotheses reaching from the formal representation of scientific knowledge to the properties of provably correct algorithms for human-machine pursuit of scientific data. The geoscientific goal is to produce the first comprehensive, time varying maps of soil strength with co-located soil moisture composition and size over the course of rainfall events in a natural landscape. The cognitive science goal is to develop a formalized representation of the cognitive processes underlying field data collection and interpretation that is simultaneously suitable to underlie robotic field assistance algorithms while at the same time advancing the study of human perceptual interpolation and reasoning. The robotics goal is to achieve a provably correct architecture for generating from formalized human task specification a chain of safe, stable online automated legged gait transitions on complex broken terrain that subserve the geoscientist data collection objectives. Two different families of legged robots with a variety of perceptual and geoscientific instrumentation suites will be deployed over natural hillsides under investigation by human geoscientists. Field performance of the resulting human-robot teams will be evaluated according to criteria assessing the degree of robotic mobility and autonomy, the quality and reliability of the resulting geoscientific measurements, and the impact of the collection process on the sampled environment. Advances in legged robot mechanics and intelligent control have brought the field to a threshold where the next major challenges for autonomous mobility can only be formulated and engaged with respect to suites of abstract but formal tasks relative to unstructured environments against which the appropriateness and success of autonomously generated, goal-directed motor behaviors can be precisely measured.
Robots endowed with even the rudiments of understanding what measurements are needed where and when by scientists, in order to test their hypotheses, would deepen our insight into the structure of human cognition. They would also open the way toward collecting massive amounts of data at presently unachievably fine spatiotemporal scales, potentially transforming the theoretical and empirical foundations of geoscience.