9:20 - 9:55: Kamal Sarabandi
Advanced Indoor Wave Propagation Model for Near Ground Ad Hoc Networks
K. Sarabandi, F. Dogfu , M. Thiel
Radiation Laboratory
Department of Electrical Engineering and Computer Science
University of Michigan
Abstract:
Communication among ad hoc nodes of an ensemble of autonomous robots is essential for their collaborative operation. The environment envisioned for operation of such robotic platforms includes complex indoor setting where the robot's transceivers are very close to the ground. If one wants to establish reliable communication link between nodes in an indoor environment, a thorough understanding of the indoor propagation is essential. While it is possible to measure the indoor channel and draw conclusions from the measurement results, such endeavors is rather time consuming and expensive. Moreover, it is difficult to extrapolate the results for other building layouts and different locations of transmitters and receivers. On the other hand accurate simulation wave propagation models give more insight to the underlying wave propagation phenomena, especially if the nodes are located near-ground.
This talk will introduce advanced methods for the modeling of indoor wave propagation. Starting from the standard method for indoor wave propagation, ray-tracing based on geometrical optics, two advanced methods are described and explained, a hybrid method that can include inhomogeneous walls and a physical optics method that can accurately capture near-ground propagation effects. It will be pointed out how the two methods can improve the accuracy of the coverage prediction compared to ray-tracing. Furthermore, the effects of the inhomogeneous walls and near-ground node placement on the coverage prediction indoors are quantified and conclusions for indoor communication with near ground nodes are drawn.
9:55 - 10:30: John S. Baras
Component-based Architectures and Synthesis for Collaborating Groups of Small Autonomous Platforms
J. S. Baras
Institute for Systems Research
University of Maryland, College Park
Abstract:
We consider collaborative autonomous control and operation of groups of small autonomous vehicles. We describe the constraints that “small” imposes on such problems with emphasis on energy, communication, memory and computation. We describe our approach and results that emphasize opportunistic communications, event-driven control, multi-modal collaborative sensing, perception-learning-cognition-learning, distributed real-time operation. Our approach follows a quantitative systems engineering methodology that starts from the system requirements and proceeds iteratively in developing system architectures and performing system tradeoffs. Several innovative design principles emerge that are unique to groups of small platforms. We also highlight several results on fundamental limitations of system performance that underlie these principles. Modularity and module adaptability and exchange emerge as fundamental principles, albeit in some new and unexpected ways. This new view of modularity is supported by examples in biology. We illustrate the results via a simple example of a small group of hybrid (ground and air) small platforms that perform collaborative tasks in a constrained in-building environment.
10:45 - 11:20: Nader Motee
Multi-Vehicle Path-Planning with Spatio-Temporal Constraints
N. Motee
Control and Dynamical Systems
California Institute of Technology
Abstract:
In this talk we will present a new framework for path planning for multiple vehicles with curvature constraints in presence of obstacles. The objective is to find multiple fixed length paths for multiple vehicles with the following properties: (i) bounded curvature (ii) obstacle avoidant (iii) collision free. Our approach is based on polygonal approximation of a continuous curve. We formulate an arbitrarily fine relaxation of the path planning problem as a nonconvex feasibility optimization problem. The constraints in the optimization problem are reformulated as non smooth forces in an elastic multiparticle system. We show that the trajectories of the nonsmooth dynamical system always converge to some equilibria that correspond to the set of feasible solutions of the relaxed problem. The proposed framework can handle more complex mission scenarios for multiple vehicles such as rendezvous while avoiding obstacles. Finally, we will present some connection between our approach and exact penalty methods for nonlinear programming problems.
11:20 - 11:55: Nathan Michael
Formation Control for a Team of Aerial Robots
N. Michael
GRASP Laboratory
University of Pennsylvania
Abstract:
We discuss the development and evaluation of a formation control strategy for a team of aerial robots. The approach is invariant to the number of robots in the team and allows for the control of group motions without specific knowledge of the full system state. The formation control law requires access to global state information that describes the group pose and shape and local information acquired directly from neighboring robots via communication. We propose the use of distributed algorithms to estimate the group descriptors and study the application of the distributed algorithm on a team of aerial robots. The experimental evaluation of these ideas leads us to argue that technological constraints such as limited bandwidth and time delays ultimately dictate the effective size of the team of robots in practice.
13:25 - 14:00: Andreas Breitenmoser, Davide Scaramuzza
Distributed Coverage Under Various Constraints for Inspection and Flying Robots
A. Breitenmoser, D. Scaramuzza
Autonomous System Lab
Swiss Federal Institute of Technology (ETH), Zurich
Abstract:
The talk outlines various constraints that arise when a robot team is to be deployed in real-world scenarios. Having our focus on multi-robot coverage, a short overview of the class of Voronoi coverage control algorithms is given. Recent work on distributed coverage in non-convex environments with a team of robots is presented. The talk is motivated by applications taken from robotic inspection of industrial plants, and surveillance of buildings and public areas in the context of the European project sFly (Swarm of Micro Flying Robots).
14:00 - 14:35: Nicholas Roy
Planning Under Uncertainty: From Single Agents to Many Agents
N. Roy
Robust Robotics Group
Massachusetts Institute of Technology
Abstract:
Planning algorithms for domains with stochastic actions and imperfect state information have historically struggled in multi-agent domains. The state space suffers a combinatorial explosion, which can make simply representing the optimal solution intractable, never mind finding the solution. However, recent work in factored models, efficient search techniques and decentralized approaches suggest that in many cases, multi-agent planning under uncertainty can be performed tractably. I will discuss some recent results from across the planning community and how they may be applicable to teams of small robots.
14:35 - 15:10: Kostas Daniilidis
Algebraic Solutions for Localization
O. Naroditsky, K. Daniilidis
GRASP Laboratory
University Pennsylvania
Abstract:
Micro-autonomous systems with limited power and hence computational constraints necessitate algorithms that can efficiently handle the data association problem and provide compact solutions for estimation. Algorithms based on algebraic geometry can exactly predict the number of solutions and can be used for minimal data samplers like RANSAC solutions for data association. We present the basic framework, applicable in several localization problems, and delve into detail for the problem of IMU supported visual odometry.
15:10 - 15:45: Stergios I. Roumeliotis
Distributed Estimation and Active Sensing with Mobile Robot Networks
S. I. Roumeliotis
Department of Computer Science and Engineering
University of Minnesota
Abstract:
In recent years, research on mobile robotics has addressed challenging estimation problems such as localization, mapping, and tracking. In most cases, the emphasis has been on how to design systems and algorithms that can perform localization, for example, in real-time within large environments and with a variety of sensors. Previous work, however, has barely begun to address important issues such as these arising from communication constraints and their impact on estimation accuracy, from the sensors' mobility and effect on the information acquired, and from the linear approximations of nonlinear process/sensor models used in practice and their impact on estimator consistency. Considering these limitations, the focus of our research has been on developing distributed estimation algorithms that explicitly consider various forms of communication limitations and provide robustness against inconsistencies due to the effect of linearization and local minima. Moreover, we have developed active sensing algorithms for determining the optimal trajectories that robots should follow in order to gather the most informative measurements for the estimated process.