There is a longstanding vision in academia: a team of unmanned vehicles cooperatively learning
and adapting in harsh unknown environments to achieve a common goal. Since the task has to
be done in a networked manner, one fundamental problem that arises in such networks is how
to maintain proper connectivity and ensure a robust flow of information. Understanding this is
crucial to realizing the multi-agent cooperative vision.
This necessitates a proper merger of communication and control issues. How much connectivity,
for instance, is needed to achieve a given collaborative goal and how can the robots enable it
through the control of motion? Wireless links can be sporadic in nature, i.e. a robot can slightly
move and lose connectivity. Thus, path planning of the robots directly affect their connectivity,
necessitating a joint optimization of motion control and communication issues. Properly
incorporating realistic communication models into the control theoretic framework can thus
bring us a step closer to realizing the vision of a team of unmanned vehicles carrying out a task
in a collaborative manner, as does tackling issues lying at this intersection of control and
Communication and robotics have traditionally enjoyed parallel and non-intersecting research.
While this intersection has seen some activity and progress in recent years, it is still relatively
unexplored and there are several interesting questions and challenges left unanswered, and
more to be discovered. The goal of this workshop is to provide a holistic view of the progress
made so far at this intersection of communication and robotics, highlighting recent results, while
also placing the spotlight on fundamental issues and challenges that arise.
10:15 - 10:30 am: Introduction and Motivation *
Dr. Yasamin Mostofi and Dr. Yorai Wardi
10:30 - 11:00 am: Talk 1
Speaker: Dr. Michael Zavlanos - Duke University
Title: Intermittent Communication Control in Mobile Robot Networks
Wireless communication is known to play a pivotal role in enabling teams of robots to successfully accomplish global coordinated tasks. For this reason, in recent years, there has been a large amount of work focused on designing controllers that ensure point-to-point or end-to-end network connectivity for all time. Available approaches rely on either graphs to model inter-robot communication or on more realistic wireless communication models that consider path loss, shadowing, and multipath fading as well as optimal routing decisions for desired information rates. Nevertheless, due to the uncertainty in the wireless channel, it is often impossible to ensure all-time connectivity in practice. Moreover, all these methods severely restrict the robots from accomplishing their tasks, as motion planning is always restricted by connectivity constraints on the network. Instead, a much preferred solution is to enable robots to communicate in an intermittent fashion, and operate in disconnect mode the rest of the time. While in disconnect mode, the robots can accomplish their tasks free of communication constraints. A great challenge in designing intermittent communication protocols is to determine which robots should communicate, where, and, more importantly, when. The goal is to ensure that connectivity of the dynamic communication network is ensured over time, infinitely often. In this talk we show that intermittent connectivity can be captured by a global Linear Temporal Logic (LTL) formula that forces robots to meet infinitely often at designated rendezvous points, and propose a novel technique to approximately decompose the global LTL formula into local ones, significantly increasing scalability of our method. We also show that the resulting controllers can be executed in an asynchronous fashion, enabling the system to handle delays and uncertainty, e.g., in the arrival times at the meeting locations. Finally, we introduce high-level robot tasks along with the communication events, and discuss theoretical guarantees, numerical simulations, and experimental results that illustrate our distributed approach.
11:05 - 11:35 am: Talk 2
Speaker: Dr. Magnus Egerstedt - Georgia Institute of Technology
Title: Constraint-Based (Communication Aware) Autonomy On-Demand
When designing coordinated control algorithms for achieving team-level objectives, safety and networking constraints, as well as power and communication considerations, are either explicitly taken into account already at the design stage, thereby making the design task significantly more complex, or added at a later stage. In this talk, we show how one can incorporate these types of lower-level considerations as constraints using Control Barrier Functions in a provably, minimally invasive manner, i.e., in such a way that the team-level objectives are respected as much as possible. Moreover, one can encode heterogeneous capabilities and objectives through these types of constraints as well, resulting in not only a richer set of behaviors but also in improved performance as compared to homogeneous formulations. In fact, in community ecology, richness of behavior is largely derived from constraints, such as scarcity/abundance of food sources or predators, and the talk will make formal connections to these ecological principles as well as deploy the developed framework on a team of mobile robots.
11:35 am - 1:00 pm: Lunch Break
1:00 - 1:30 pm: Talk 3
Speaker: Dr. Yasamin Mostofi - University of California Santa Barbara
Title: Communication-Aware Control and Robotics: Opportunities and Challenges
In this talk, I focus on communication-aware robotics and human-robot networks. I will start by showing how each robot can go beyond the over-simplified but commonly-used disk model for connectivity, and realistically assess the impact of a motion decision on its link quality. By utilizing this framework, I will then show how each unmanned vehicle can best co-optimize its communication, sensing and navigation objectives under resource constraints. This co-optimized approach results in a significant performance improvement as I discuss in the talk. Finally, I show some recent results on human-robot networks that enable the robots to predict human visual performance and incorporate it in their sensing and path planning optimization.
1:35 - 2:05 pm: Talk 4
SPeaker: Dr. Urbashi Mitra - University of Southern California
Title: Collaborative detection and estimation over shared networks
Consider a robotic network whose goal is to perform a given distributed sensing task such as detecting an intruder in a security and surveillance application, or estimating abrupt changes in vital parameters of a system in real-time. When such sudden events occur, in order to overcome the intrinsic limitations of the wireless communication medium, the robotic sensor network typically needs to employ coordination algorithms to facilitate access to the channel and avoid packet collisions that may occur due to interference. However, these algorithms are usually costly to implement and cause undesirable delays, which may incurr in a potential irreversible loss.
In this talk, we will explore a new class of problems where a team of agents perform collaborative sensing tasks without the aid of channel access protocols, over a shared wireless network modeled as a collision channel. This channel model is characterized by the fact that if two or more sensors transmit packets at the same time, a collision is declared and information is lost. First, we will show the existence of jointly optimal communication policies with a particular threshold structure for the problem of decentralized estimation of a continuous random vector over the collision channel in the mean squared sense. In addition to being easily implementable, such threshold policies can be designed using an efficient variation of the Lloyd-Max algorithm. Then, we will address the decentralized detection of discrete events with a minimum probability of error criterion. In this problem formulation, we will show the existence of a jointly optimal policy where the sensors transmit all except the most likely of their observations. In both cases we argue that the optimal policies exploit the actions to transmit or to remain silent as a form of implicit communication. We will conclude with a set of open questions and applications to teams of mobile sensors.
2:10 - 2:40 pm: Talk 5
Speaker: Dr. Athina Petropulu - Rutgers University
Title: Enhancing QoS in Beamforming Networks: Mobile Beamformers and Optimal Motion Policies
Distributed, networked communication systems, such as relay beamforming networks (e.g., Amplify and Forward (AF)), are typically designed without considering how the positions of the respective nodes might affect the quality of the communication. That is, network nodes are either assumed to be stationary in space, or, if some of them are moving while communicating, their trajectories are assumed to be independent of the respective communication task. However, in most cases, the Channel State Information (CSI) observed by each network node, per channel use is both spatially and temporally correlated. One could then ask whether the performance of the communication system could be improved by (predictively) controlling the positions of the network nodes (e.g., the relays), based on observing causal CSI and exploiting the spatiotemporal dependencies of the communication medium. In this talk, we address the problem of enhancing Quality-of-Service (QoS) in power constrained, mobile AF relay beamforming networks, by optimally exploiting relay mobility. We consider a time slotted system, where the relays update their positions before the beginning of each time slot. Adopting a spatiotemporal stochastic field model of the wireless channel, we propose a novel 2-stage stochastic programming formulation for specifying the relay positions at each time slot, such that the QoS of the network is maximized on average, based on causal CSI and under a total relay transmit power budget. Via the Method of Statistical Differentials, the motion control problem considered is shown to be approximately equivalent to a set of simple subproblems, which can be solved in a distributed fashion, one at each relay. It is also shown that, under mild, natural assumptions on the stochastic nature of the channel model considered, the proposed 2-stage framework exhibits a striking property: The average network QoS increases across times slots. Numerical simulations are presented, corroborating the efficacy of the proposed approach and confirming its properties.
2:45 - 3:00 pm: Coffee Break
3:00 - 3:30 pm: Talk 6
Speaker: Dr. Yorai Wardi - Georgia Institute of Technology
Title: Balancing transmission power with motion power in mobile networks: an optimal control perspective
This talk presents the problem of optimally balancing transmission power (or energy) vs. motion power in a network of mobile robots. The robots are tasked with the transmission of data while in motion, and the problem is to compute their scheduled motion trajectories and other parameters (such as transmission rates) that minimize the total energy. This problem is cast in the framework of optimal control, and a specialized algorithm for its solution is presented. The algorithm is designed for decentralized implementations in real-time environments, and it displays fast convergence towards optimal solutions. Illustrative examples will be presented, and several directions for future research will be discussed.
3:35 - 4:05 pm: Talk 7
Speaker: Dr. Bhaskar Krishnamachari - University of Southern California
Title: A Networking Perspective on Distributed Robotics
In many distributed multi-robot applications, the robots will need to communicate via radios with each other. Such a system, including robotic relay nodes, provides unique challenges and opportunities from a networking perspective. The ability to control the location and motion of individual relay nodes brings a new dimension to wireless protocols which today are designed assuming not only uncontrollable but also unpredictable. I will present several case studies from work at the USC Autonomous Networks Research Group addressing the emerging new domain of robotic wireless networks. These include algorithms and protocols for rapid robotic network deployments in unknown obstructed environments, flow-adaptive network topology re-configuration, and backpressure-based robotic message ferrying.
4:10 - 4:40 pm: Talk 8
Speaker: Dr. Ceyhun Eksin - Georgia Tech
Title: Decentralized fictitious play-type algorithms for multi-agent systems with uncertainty
We consider multi-agent systems deployed in unknown environments assigned to collectively achieve a global objective. The global objective depends on the actions of each agent and the environment. Agents have different and time-varying beliefs on the environment. As a consequence, they cannot agree on the global objective. In this setting, we first define the optimal agent behavior using the concept of Bayesian Nash equilibria where each agent reasons about what the actions of other agents would be given only local observations. We then present a family of dynamics based on the fictitious play algorithm that allows agents to repeatedly make sub-optimal decisions while receiving local messages from their neighbors in the communication network. If agents move towards a common belief on the environment, we establish almost sure convergence of the algorithms to a Nash equilibrium of the game with utility functions given by the expectation of the global objective computed with respect to the common belief. We further study fully-distributed instances of the proposed dynamics that overcome the computational complexity and informational demands of Bayesian Nash equilibria. Finally, we corroborate these results with numerical experiments and implementation on a multi-robot test-bed for the target covering problem.
4:45 - 5:30 pm: Panel Discussion
* Coffee and snacks will be served at 10 am before the workshop begins.