DARPA YIA 2014: Specification and Control of Customizable Multi-Robot Systems for Distributed Sensing and Cooperative Manipulation Project Summary P.I.: Spring Berman, Ph.D. Unmanned systems in the military have proven to be productive and costeffective assets in missions that are characterized as dull, dirty, and dangerous, including surveillance, reconnaissance, target acquisition, transport, explosive ordnance disposal, mine clearance, and security inspections. Systems comprised of multiple expendable robots that coordinate with each other, perform low-level functions autonomously, and receive high-level instructions from human supervisors have the potential to perform these tasks on large spatial and temporal scales, greatly reduce human workload, remove more military personnel from danger, and perform riskier missions than manned assets. The production and deployment of multi-robot systems in practice is nearing feasibility due to recent advances in computing, sensing, actuation, power, communication, and control technologies. In the last few years, the miniaturization of these technologies has led to numerous robotic platforms that are designed to act in collectives, including micro quadrotors, flapping-wing micro aerial vehicles, and multi-legged ambulatory robots. These small, inexpensive platforms are designed to accommodate a variety of sensors and employ diverse, often biologicallyinspired locomotive capabilities. Coordinated groups of such platforms could potentially conduct a range of sensing and manipulation tasks very efficiently, unobstrusively, and at a relatively low cost. In addition, current research on 3D-printing technology will enable military personnel to quickly fabricate custom parts, electromechanical systems, and even entire robots for target applications without the need for expert knowledge. Building on these recent and ongoing technological developments, the proposed research will enable non-expert users to automatically determine the physical and functional composition of a customizable multi-robot system that will optimize a specified performance metric for a particular mission. This research seeks to overcome several outstanding challenges in the design and control of multi-robot systems outside of predictable, structured environments. Our system design approach will determine the sets of sensing and manipulation capabilities and the population of robots with each capability set that will optimize metrics such as time to mission completion, extent of sensor coverage, total power consumed, and system cost. Our approach to robot controller synthesis is scalable with the number of robots, robust to communications degradation, agnostic to prior data and GPS signal, and utilizes verification and validation methods for non-deterministic behaviors such as random encounters with environmental features. It will produce target robot allocation and manipulation behaviors using only local sensing, local communication, and common broadcast information. The proposed multi-robot system design and control approach will consist of a novel integration of tools and techniques from stochastic process theory, chemical reaction theory, dynamical systems and hybrid systems analysis, feedback control theory, and optimization. The approach will employ stochastic, deterministic, and hybrid stochastic-deterministic models at different levels of abstraction that describe the robots roles, task transitions, motion, and manipulation dynamics. At the highest abstraction level, we will use ordinary and partial differential equations to model the system population dynamics. Such models have the advantage of being amenable to rigorous analysis, control, and optimization techniques that can provide theoretical guarantees on system performance; however, they require knowledge of the parameters of the environment in which the system operates. We will circumvent this limitation by using the models to derive stochastic robot control policies that depend predictably on parameters of the robots but are independent of the environment. This innovation will build on our recently published work on top-down controller synthesis for stochastic boundary coverage by multi-robot systems. To derive the control policies, we will apply a range of techniques to the differential equation models, including feedback control of linear systems, multi-affine systems, and piecewise-multi-affine hybrid systems on polytopes, and methods for convex optimization and stochastic optimization. We will experimentally validate our models and control strategies on a multi-robot testbed consisting of 100 small, inexpensive ground robots that will be designed and fabricated by members of the PIs lab. Experiments will be conducted on sensor coverage of both open and closed boundaries of various geometries and collective transport of payloads through environments with and without obstacles.
|Effective start/end date||9/15/14 → 9/14/16|
- DOD: Defense Advanced Research Projects Agency (DARPA): $743,248.00