Matthew Travers - US grants

Sub-millimeter scale cyber-physical systems will have a major impact on future applications.  For example, targeted drug delivery or materials conveyance for micro-scale construction are two important application areas on which small-scale systems will advance the current state of the art.  However, conventional actuator, sensor, and computational units are generally not available at extremely small scales.  This project thus explores the relationships between novel microfabrication, system design for articulated locomotion, and active control of micro, cyber-physical systems.  More specifically, this project develops a common analytical framework to understand, express, and reason about the connections, as well as demonstrate on a novel problem, the benefits of self-propelled articulated micro-scale devices.

The project is developing elasto-magnetic filaments formed by linked ferromagnetic beads.  These filaments can serve as the basis for functionalized structures, employing protein-coatings, that are flexible and controllable through actively manipulated distributions of magnetic dipole moments.  This approach uses dual laser polymerization to construct templates that enable the magnetization profile of chains composed by single micron diameter ferromagnetic spheres, bonded by DNA origami strands, to be actively programmed.  These elasto-magnetic bodies are then articulated by changes in an externally applied magnetic field, i.e., when subjected to a constant but oscillating weak magnetic field, the local alignment of dipole moments to the field will actively "actuate" the systems.  The analysis, based on a geometric framework, will determine the optimal distribution of magnetization profiles across the filaments; thereby linking fabrication to analysis and the geometry underlying locomotion in dissipative fluids to novel maneuvering capabilities.  Guided by this framework, as a demonstration, microrobots with these magnetized bodies will be designed to achieve specific locomotion objectives in sufficient numbers to be made to move purposefully in uncertain environments.

Unconventional, underactuated robots, such as humanoids or legged platforms more broadly, offer the potential to move through and perform work in constrained, three-dimensional environments that are currently inaccessible to existing autonomous agents. However, this potential has been largely unrealized because it is difficult to reliably adapt the behaviors of these platforms to account for the changing and uncertain task and environmental conditions in the "real world." Although many of the fundamental principles that govern contemporary task and motion planning techniques are applicable across different platforms, the practical implementation of these principles has been largely platform specific. In contrast, this project will adopt a probabilistic planning framework which learns common structure for the motion patterns of different platforms performing related tasks, then uses this structure to generate generalized, inherently platform independent, motion primitives. At runtime, the primitives will be grounded and adapted where necessary to specific robot models given local task and environmental conditions. The primary benefit of this project will be an increase in the utility of autonomous platforms for tasks such as urban search and rescue, industrial inspection, and planetary exploration. The analytical techniques that will be developed will have further impacts on locomotion science and learning-based approaches to motion coordination. The PIs will additionally be involved with K-12 outreach involving robot demonstrations at FIRST Robotics Competitions and the Rochester Museum and Science Center.

This project will specifically address fundamental limitations in the tractability of real-time task and motion planning for underactuated robots over diverse objectives and distributions of environmental conditions. Probabilistic models will be developed to efficiently reason over and adapt the nominal behaviors of different highly-articulated, underactuated robots. The behavioral inference will make it possible to 1) select appropriate pre-existing behaviors (developed over the course of the project) where relevant, 2) use novel combinations of nominal behaviors to form compound, task-specific behaviors, and 3) leverage similar, but not necessarily the same, kinematic structure across heterogeneous platforms to transfer behaviors between them. To ensure the success of the practical, online implementation of the developed models, the PIs will develop algorithms that combine probabilistic inference, nonlinear dimensionality reduction, and dynamic movement primitives to produce a novel combination of efficient motion generation and robust online adaptation. In addition to varying task and environmental conditions, the adaptability of the probabilistic models to changes in the internal kinematics and dynamics of robot platforms, such as those that would arise from degraded motor performance or structural failures of joints or entire limbs, will also be explored. The models will be trained and validated using a combination of simulation and experimental results on two physical platforms: the Carnegie Mellon Hexapod and the Robotis OP2. Furthermore, the PIs will develop software tools and release open-source products related to generalizable probabilistic models for motion adaptation of underactuated systems.

Future micro-scale robotic devices will be integrated in broad aspects of human life, having explicit impact on applications as diverse as non-invasive surgical procedures to advanced electronics manufacturing. However, at very small physical scales it is essentially impossible to manipulate, in a conventional sense, the physical components that are necessary to construct mechanical structures. In addition, it is also very difficult to see or otherwise sense with fidelity mechanisms that are formed at extremely small scales, e.g., the nano-scale. This inability to easily sense nano-scale mechanisms makes it difficult to gather data on the efficacy of different assembly processes or collect feedback data that would be required to control active mechanisms, i.e., to actuate them to induce some desired motion. Therefore, to address these difficulties, this work proposes to develop a synergistic framework that combines ideas related to contemporary top-down and bottom-up manufacturing processes with those from the machine learning (ML), artificial intelligence (AI), and robotics communities to address what we see as the largest current barriers to the practical deployment of future nano-scale robotic systems: 1) manipulating components for assembly, 2) the availability of low-cost, readily available sensing, and 3) actuating the mechanisms once formed. In addition, we propose to make broader contributions to the research community by establishing a framework for archiving multimodal data about nanostructure formation statistics that will be shared with and ideally added to by other researchers. Lastly, from an educational perspective, the PIs have already begun to educate middle school students about artificial intelligence and DNA nanotechnology and intend we further these efforts by introducing a novel macroscale model that can serve as an interactive activity for teaching K-12 students more about the confluence of DNA nanotechnology and artificial intelligence.

This work develops a novel framework that optimizes the outcome of physical processes wherein modular nanoscale robots self-assemble from a set of nano submodules. The main contribution of the framework to be developed is to reduce the uncertainty in large-scale self-assembly processes wherein the objective is to create nanoscale superstructures with specific designs. We propose to use DNA origami to create modular components in a nanoscale test bed because DNA origami is an excellent tool for forming different geometric constructs, e.g., a honeycomb-like truss, with subnanometer precision. We intend to use a graph neural network framework to model complex, large-scale self-assembly processes as distributions over a discrete space of modular DNA superstructures that are represented using graphs. We hypothesize that this approach will allow us to optimize the process conditions during the manufacturing trials, such as the number of unique connections between components, thereby maximizing the yield of desired superstructures. Our key contributions include 1) learn to map low dimensional characterization data to a graph-based representation of the corresponding superstructure populations; 2) generate probabilistic graphs that represent the distribution of superstructures formed for an arbitrary set of manufacturing conditions; and, 3) apply optimization techniques to our generative model to find the optimal manufacturing conditions for maximizing the yield of a desired superstructure. In addition, Taylor and Travers plan to leverage their ongoing collaboration that focuses on using global stimuli like magnetic actuation to induce motion in systems constructed using magnetic micro- and nanoparticles to perform preliminary motion studies that will be conducted using chemical and magnetic field actuation.

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.