Focus. My goal is to build resilient heterogeneous robotic teams and systems by ensuring the integrity of interaction, collaboration and coordination mechanisms, and by providing resilience to environmental disruptions and misleading sensory information. I take a holistic approach, by jointly considering (i) abstractions of robotic interactions, (ii) precautionary control strategies, and (iii) resilient inference algorithms.
Motivation. We are witnessing a profusion of networked robotic platforms with distinct features and unique capabilities. To exploit the diversity of such robotic networks, we are contriving ecosystems of tightly interconnected and interdependent heterogeneous entities. However, as connections are established, information is shared, and dependencies are created, these systems give rise to new vulnerabilities and threats. If our systems are to succeed, they must be built to resist disruptive events and malicious intentions. I argue that resilience needs to become a central engineering paradigm.
Privacy & Resilience in Networked Robotic Systems
A quantitative privacy model. We are interested in securing the operation of robotic networks composed of heterogeneous agents. Since any given robot type plays a role that may be critical in guaranteeing continuous and failure-free operation of the system, it is beneficial to hide information that reveals the individual robot types and, thus, their roles. We propose a method that quantifies how easy it is for an adversary to identify the type of any of the robots, based on an outside observation of the system’s behavior. We draw from the theory of differential privacy, and develop an analytical model of the information leakage. This model allows us to analyze the privacy of the system, as its parameters vary.
[SI 2017] [DARS 2016] [ANTS 2016]
Resilient networks and control. All cooperative control algorithms for robot teams rely on the assumption that all entities are, in fact, cooperative. This, however, cannot be generally guaranteed, as robots break, are compromised, or fail to adequately process and interpret sensor information. We consider two primary ways of providing resilience, (i) by building resilient robot formations, and (ii) by providing resilient control strategies. Our methods are underpinned by recent results in network science that define the notion of robust communication graphs. We are the first to apply these concepts to the domain of robotics by considering physically embedded multi-agent systems, with constrained communication radii and dynamic behaviors. Thus far, we have considered applications to formation control, flocking, and monitoring/coverage tasks.
[RAL 2016-a] [RAL 2017-b] [ACC 2017] [DARS 2016-b]
Control of Heterogeneous Robot Swarms
The impact of diversity. As we aspire to solve increasingly complex problems, it becomes ever more difficult to embed all necessary capabilities into one single robot type. Therefore, we distribute distinct capabilities among robot team members. During this process, heterogeneity becomes a design feature. The question is, then, how to best design such systems so that the resulting performance is optimized. We propose a framework that represents diversity explicitly: our metric defines the notions of minspecies, eigenspecies, and coverspecies, i.e., the minimum set of species (types) that are required to achieve a particular goal. The metric enables a quantitative analysis of the relation between diversity and performance.
[TRO 2016] [ICRA 2016]
Control policies. We consider the problem of distributing a large group of heterogeneous robots among a set of tasks that require specialized capabilities (traits) in order to be completed. Our control solution implicitly solves the combinatorial problem of distributing the right number of robots of a given species to the right tasks. To find the optimal control policy, we develop a method that is fully scalable with respect to the number of robots, number of species and number of traits. Building on this result, we propose a real-time optimization method that enables an online adaptation of transition rates as a function of the state of the current robot distribution.
[TRO 2016] [BICT 2015] [Acta 2016]
Localization in Networked Robotic Systems
Ultra-wideband (UWB) localization. Ultra-wideband (UWB) localization is a recent technology that promises to outperform other indoor localization methods. Yet, non-line-of-sight scenarios cause large biases in the signal propagation times, which lead to large localization errors. This work addresses the peculiarities of UWB signal propagation with a closed-form time-difference-of-arrival (TDOA) measurement model, and represents the first such measurement model for mobile robot localization. The application of the model enables indoor localization for miniature mobile targets with errors in the order of a few centimeters.
[IPSN 2017] [IJRR 2014] [ISER 2013] [ICRA 2012] [IPIN 2011] [IPIN 2010]
Collaborative localization. Localization is an enabling technology, and a prerequisite for a wide range of tasks. We provide a scalable solution to the localization problem for fully decentralized, large-scale multi-robot systems. The resulting algorithm is based on particle filtering, and two additional algorithmic components (reciprocal sampling and clustering) that enable robust and efficient localization. The result is a low-cost collaborative localization algorithm that is fully scalable with respect to the number of robots in the system. We develop a framework that combines the collaborative localization method (based on relative range and bearing measurements) with absolute UWB positioning. Results show that the integration of cheap relative positioning hardware represents a cost-efficient way of improving the accuracy of any absolute positioning system using time-of-arrival based measurements.
[Chapter 2012] [ICRA 2012] [IROS 2011] [IROS 2011b]
Modeling Distributed Robotic Systems
Spatial modeling. A large body of work in the modeling of robotic swarms relies on the assumption that the system is well-mixed, and that non-spatial metrics can be applied. This assumption, however, is often violated. Structured environments and ambient conditions may induce spatial drift, and hence, cannot be modeled by non-spatial metrics. This work presents the first spatial modeling framework able to combine macroscopic models (that capture the behavioral states of robots) with diffusion models (that capture the spatial distributions as a function of time). The framework is validated by experiments with real robots; the results show that more accurate predictions on system performance can be made.
Sensor network topologies. The efficiency of distributed sensor networks depends on an optimal trade-off between the usage of resources and data quality. This work addresses the problem of optimizing this trade-off in a self-configured distributed sensor network, with respect to a user-defined objective function. The novelty of this work lies in the implementation of a quadtree network topology that is maintained through local (node-level) decisions. An innovative analytical model of the network shows very good correspondence with real-life experiments. This result enables the optimization of an objective function that defines the trade-off based on user needs.
[ICRA 2010] [IROS 2011c]