Gregory S. Chirikjian - US grants

9357738 Chirikjian This award funds the first-year base amount of a five-year NSF Young Investigator Award. The research goals consist of two separate projects. First, the kinematics, motion planning, and design of mechanisms with `binary actuators' will be investigated. Robots with binary (on-off) actuators have a finite number of states. Major benefits of binary actuated manipulators are that they can be operated without extensive feedback control, task repeatability is very high, and two-state actuators are generally very inexpensive, thus resulting in low cost robots for industrial use. Second, the design and coordination of `metamorphic' robotic systems for use in unknown environments will be investigated. A metamorphic robot is a collection of independent modules each of which has the ability to locomote over adjacent modules. The morphology of the collection of modules changes as a function of the collective motion. Such self-reconfigurability potentially gives metamorphic robots unique abilities. For instance, modules can `flow' into crevasses unreachable by fixed morphology robots, or grow to form structures which envelop objects and form stable grasps.

The research consists of two separate projects, in discrete robotic system design and discrete motion planning. First, the kinematics, motion planning, and design of robotic mechanisms with binary actuators are investigated. Robots with binary (on-off) actuators have a finite number of states. Major benefits of binary-actuated manipulators are that they can be operated without extensive feedback control, task repeatability is very high, and two-state actuators are generally very inexpensive, thus resulting in reliable low cost robots for industrial use. Second, the design and coordination of`metamorphic' robotic systems for use in unknown environments are investigated. A metamorphic robot is a collection of independent modules each of which occupies a space in a lattice, and has the ability to locomote over adjacent modules. The morphology of the collection of modules changes as a function of the collective motion. Such self-reconfigurability gives metamorphic robots unique abilities. For instance, modules can `flow' into crevasses unreachable by fixed morphology robots, or grow to form structures which envelop objects and form stable grasps.

The field of service robotics holds the potential to increase the quality of life of millions by removing people from situations that are dirty, dull, and dangerous. However, current robots are far too expensive, have limited task flexibility, and lack the robustness required for use in service applications. A paradigm shift is required if service robots are to ever be as ubiquitous as personal computers. The goal of this research is to develop a "toolbox" containing simple actuators and structural elements together with a design methodology which governs how the components should be assembled for specified service tasks. The cornerstone of this approach is the use of binary (two-state) actuators such as pneumatic cylinders because of their high payload to weight ratio and very low cost. When used in combination with passive damping elements (dashpots) such actuators provide smooth and well-behaved motion between binary states with very little backlash. Within a specified set of tasks the robot constructed from this toolbox is essentially a dedicated machine. The dedicated nature of the resulting machine reduces the cost requirements and increases the reliability of the robot as compared to general purpose robots.

The proposed work addresses a variety of issues related to motion planning of complex systems: development and application of statistical methods to reduce the effect of the "course of dimensionality" in configuration-based motion planning; constructing stochastic models of mechanical systems to account for noise in low-level sensing and actuation and its effect on high-level planning; coordination of multi-robot swarms and their biological analogues. The obtained results will be tested on holonomic as well as non-holonomic systems, and also applied to studying other complex systems, such as protein molecules.

This project is a multidisciplinary, international collaborative research project aimed at developing a fully automated robotic system for on demand and batch scanning of print materials and an open-source software framework that will begin with a printed work and end with digital images, text and musical content suitable for digital libraries.

DESCRIPTION (provided by applicant): This is a proposal to bring techniques from group theory and (noncommutative) harmonic analysis to bear on computational aspects of structural biology. Novel algorithms will be developed and implemented in order to address the following problems: de novo determination of protein structure from unassigned residual dipolar couplings without a prior knowledge of the Saupe alignment tensor; rapid minimization of the "rotation function" for molecular replacement of multi-domain proteins; de novo determination of electron densities from projections in cryo-electron-microscopy. Our unified approach casts these as minimization problems and fast functional evaluations on finite and Lie groups, for which we will need to generalize methods such as gradient descent and FFTs to the group-valued setting. Our team combines expertise in mathematics, engineering, and biology necessary to make progress in this highly interdisciplinary subject. The problem of protein structure determination is central to the understanding of protein function and molecular design. This is absolutely critical to the progress of the health and environmental sciences in regards to efforts related to "designer drugs" - i.e., the development of targeted therapies, be they for humans, animals, or plants. Our efforts have the potential to remake high-throughput techniques by providing experimentalists with new and efficient techniques for the comparison of molecular structures. The novelty of our approach is in its focus on the use of the tools of group theory and group representation theory (harmonic analysis) in this life sciences setting. A particularly attractive aspect of this proposal is the close connection between theory and practice. Our interdisciplinary team combines mathematical, computational, biological, and engineering skills - so that it is well-poised to make progress on a problem that is inherently multidisciplinary and one that draws on each of these areas

This project develops a framework for deploying multi-robot teams on man-made structures for continual inspection, structural health monitoring, and minor repair tasks.

Within this framework, sensory information obtained by a coordinated group of mobile robots is pooled, resulting in a diagnosis of the state of the structure on which the robots roam. In addition, methodologies for endowing the robots with the ability to self-diagnose are being explored. These topics involve the integration of advanced probabilistic, geometric, and mechanics-based computations.

The results of this research are expected to enhance the robustness, lifetime, and range of applicability of robotic systems. Research results will be leveraged by collaborating with external research labs.

The robots and structures in the testbed being developed in this project, which are scaled down models of real-world systems, are ideal for student projects.
Participation by undergraduates and local high-school students on research projects in the PI's lab will serve as a model for increasing interest in engineering research among these groups.

This project, developing advanced mathematical and computational methods for the online calibration of ultrasound probes that takes into account a probabilistic version of the well-known AX = XB sensor-calibration problem that has been overlooked in the robotics and computer vision literature, will advance current capabilities in computer-integrated surgical interventions, leading to lower radiation exposure to patients and better outcomes for minimally-invasive surgery.

Ultrasound has many benefits for minimally-invasive surgical procedures, including cost, ease-of-use, and patient/doctor radiation exposure. But ultrasound images are fuzzy and require extensive training for proper use during surgical procedures. As a result, outcomes are heavily dependent on an individual surgeon's skill with the device.

Broader Impacts: Beyond the potential benefits to surgical procedures, the Laboratory for Computational Sensing and Robotics (LCSR) at JHU has an established summer program for visiting undergraduate students that will facilitate involvement of undergraduates in the proposed research. In addition, the PI continues to mentor high school students from Baltimore Polytechnic High School through research experiences both during the academic year and the summer. The hands-on and visual nature of ultrasound image acquisition together with the mathematical problems of registration and calibration make this an ideal project to introduce students to the importance of mathematics.

DESCRIPTION (provided by applicant): Due to their inherently complex nature, the architecture and motions of large macromolecular assemblies composed of rigid constituents are typically dissected using multiple techniques. While often combined on a case-by-case basis, the lack of theoretical tools to optimally integrate information from different sources is a major barrier to generating a more complete/accurate understanding of important assemblies. Herein are proposed new information fusion algorithms for these assemblies, and their associated functional motions. This extends classical information science to the case of data on the Lie group of rigid-body motions. Utilizing data from electron microscopy (EM) and small-angle X-ray scattering (SAXS) measurements, these fusion algorithms will be applied to two large biomolecular assemblies: (1) the ionotropic glutamate receptor (iGluR), and (2) the Chd1-nucleosome complex. The specific aims are as follows: SA1: To develop new information-theoretic methods based on Euclidean-group calculus and probability theory to improve fitting of macromolecular structures into EM densities and SAXS envelopes, and to perform information fusion of compatible biophysical information from different modalities to produce greater understanding than when methods are taken individually. SA2: To apply mathematically optimized models of iGluR quaternary structure to uncover physiologically relevant conformational changes inaccessible to individual experimental methods. SA3: To develop and apply new mathematical models of flexibility and ensemble dynamics of the nucleosome alone and in complex with the Chd1 chromatin remodeler using EM and SAXS, leading to a better understanding of the structure-motion-function relationship. The results will validate novel algorithms for fusing information from different experimental approaches to determine conformational changes in macromolecular complexes. If successful, these algorithms will provide new mechanistic insights into the iGluR family of ligand-gated ion channels, implicated in stroke and Alzheimer's disease, and the Chd1 remodeler, which has been linked to several types of cancer.

This INSPIRE project is jointly funded by Algorithmic Foundations in CISE/CCF, Analysis in MPS/DMS, and the NSF Office of Integrative Activities.

Knowledge of the 3-dimensional structure of protein molecules supports scientific understanding of how proteins perform their functions within cells. Structures of over 100,000 proteins in the Protein Data Bank have been determined by macromolecular x-ray crystallography: measuring the diffraction of x-ray beams from crystal composed of many symmetrically arranged copies of one or more protein molecules gives partial information (the amplitudes of the Fourier transform) that must be filled in (solving the "phase problem," often by molecular replacement -- taking phases from related molecules) to complete the 3d structure. Molecular replacement works quite well for simple single-domain proteins, but breaks down for multi-domain proteins and large complexes; one needs to explore the possible combinations of domains and their diffraction patterns as replacement candidates.

This cross-disciplinary project brings together experts in robotics and in pure mathematics to address the ''phase problem'' of macromolecular x-ray crystallography. The mathematical and computational framework developed in this project will enable many more protein structures to be solved in a less laborious way than can be done now. The project also introduces Baltimore City high school students to mathematics and molecular biophysics through unique visualization activities.

The essence of combining domains is geometric. The team can use articulated multi-rigid-body models from the field of robotics to combine rigid portions of structures, from domains with similar sequences. The relative rigid-body motions between the domains become the unknown degrees of freedom in these articulated models. Crystal packing constraints will rule out the majority of possible configurations for these domains, and reduce the otherwise high-dimensional nature of the search space. The project will develop new algebro-geometric and computational methods for rapidly discarding the large collision regions in configuration space, so that searches will focus on the remaining small-volume feasible regions in this high-dimensional search space. Computer code will be prepared integrating the resulting methods into existing molecular crystallography software packages.

Motion planning is an important part of robotics research that enables robots to move without colliding into obstacles. This project focuses on a new paradigm for motion planning in cluttered environments in which robots and obstacles are represented as unions of ellipsoids, rather than the traditional approach of using polyhedral, which makes it more efficient to calculate potential collisions during planning. In addition to the scientific goals of the project, a longstanding outreach activity with Baltimore Polytechnic High School is part of the project. Two students from this high school per year will spend their "research practicum" in the PI's laboratory during their senior years.

In terms of detailed scientific questions, this project builds on the fact that the set of rigid-body motions that cause two ellipsoids to be in collision can be parameterized in closed form, thereby facilitating the sampling of collision-free configurations outside of these sets rather than wasting computational resources on sampling, evaluating whether collisions occur, and throwing away large numbers of
samples. This is particularly important in so-called "narrow-passage" problems. This approach therefore has the potential to dramatically improve the performance of motion planning algorithms.

This award will support the organization of the annual Principal Investigators (PI) meeting for the National Robotics Initiative (NRI), which was launched in 2011. The PI meeting will bring together the community of active NRI participants to provide cross-project coordination of intellectual challenges and best practices in education, technology transfer and general outreach. This activity will also establish a repository illustrating the research ideas explored and milestones achieved by the NRI projects.

The meeting is planned for two days in October 2018 in the vicinity of Washington DC. The format will include presentations by the attending PIs of projects in their final year, a poster session with all other projects, keynote speeches, and panel discussions.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.