Leonardo Bobadilla

Controlling Wild Bodies Using Linear Temporal Logic.

There is substantial interest in controlling a group of bodies from specifications of tasks given in a high-level, human-like language. This paper proposes a methodology that creates low-level hybrid controllers that guarantee that a group of bodies execute a high-level specified task without dynamical system modeling, precise state estimation or state feedback. We do this by exploiting the wild motions of very simple bodies in an environment connected by gates which serve as the system inputs, as opposed to motors on the bodies. We present experiments using inexpensive hardware demonstrating the practical feasibility of our approach to solving tasks such as navigation, patrolling, and coverage.

Controlling wild mobile robots using virtual gates and discrete transitions

We present an approach to controlling multiple mobile robots without requiring system identification, geometric map building, localization, or state estimation. Instead, we purposely design them to execute wild motions, which means each will strike every open set infinitely often along the boundary of any connected region in which it is placed. We then divide the environment into a discrete set of regions, with borders delineated with simple markers, such as colored tape. Using simple sensor feedback, we show that complex tasks can be solved, such as patrolling, disentanglement, and basic navigation. The method is implemented in simulation and on real robots, which for many tasks are fully distributed without any mutual communication.

Manipulating Ergodic Bodies through Gentle Guidance

This paper proposes methods for achieving basic tasks such as navigation, patrolling, herding, and coverage by exploiting the wild motions of very simple bodies in the environment. Bodies move within regions that are connected by gates that enforce specific rules of passage. Common issues such as dynamical system modeling, precise state estimation, and state feedback are avoided. The method is demonstrated in a series of experiments that manipulate the flow of weasel balls (without the weasels) and Hexbug Nano vibrating bugs.

Minimalist multiple target tracking using directional sensor beams

We consider the problem of determining the paths of multiple, unpredictable moving bodies in a cluttered environment using weak detection sensors that provide simple crossing information. Each sensor is a beam that, when broken, provides the direction of the crossing (one bit) and nothing else. Using a simple network of beams, the individual paths are separated and reconstructed as well as possible, up to combinatorial information about the route taken. In this setup, simple filtering algorithms are introduced, and a low-cost hardware implementation that demonstrates the practicality of the approach is shown. The results may apply in settings such as verification of multirobot system execution, surveillance and security, and unobtrusive behavioral monitoring for wildlife and the elderly.

Combinatorial Filters: Sensor Beams, Obstacles, and Possible Paths

A problem is introduced in which a moving body (robot, human, animal, vehicle, and so on) travels among obstacles and binary detection beams that connect between obstacles or barriers. Each beam can be viewed as a virtual sensor that may have many possible alternative implementations. The task is to determine the possible body paths based only on sensor observations that each simply report that a beam crossing occurred. This is a basic filtering problem encountered in many settings, under a variety of sensing modalities. Filtering methods are presented that reconstruct the set of possible paths at three levels of resolution: (1) the possible sequences of regions (bounded by beams and obstacles) visited, (2) equivalence classes of homo-topic paths, and (3) the possible numbers of times the path winds around obstacles. In the simplest case, all beams are disjoint, distinguishable, and directed. More complex cases are then considered, allowing for any amount of beams overlapping, indistinguishability, and lack of directional information. The method was implemented in simulation. An inexpensive, low-energy, easily deployable architecture was also created which implements the beam model and validates the methods of the article with experiments.

Modeling and analyzing occupant behaviors in building energy analysis using an information space approach

Buildings account for a majority of energy consumption in the United States. One of the major factors affecting the energy performance of buildings is occupant behaviors. Decoding occupant behaviors is a key to identifying energy waste and to discovering strategies to curtail energy consumption in buildings. We propose an information space approach for automated detection and proactive monitoring of energy waste due to occupant behaviors. In this paper we present a set of filtering algorithms to capture the minimum amount of information necessary to detect wasteful states and trajectories that occupants may have, in order to pro-actively modify occupant behaviors. We also describe and implement a sensor network consisting of inexpensive distance, light, temperature sensors and electricity consumption monitors utilized in order capture data related to occupancy behaviors. By keeping count of the number occupants and energy expenditures in different regions of a building, we accurately estimate how occupancy behavior is affecting energy use, in a non-invasive way. Furthermore, we present a methodology to pro-actively eliminate energy expenditure by calculating a score associated with occupants in different regions. This score will be used to suggest policies to users or facility managers to help reduce energy costs related to occupancy behaviors.

A genetic word clustering algorithm

In this work, a genetic word clustering algorithm, that classifies words present in the phrases of a linguistic corpus, is proposed. The underlying goal of word classification is to build a good probabilistic model of the language defined by the phrases in the corpus. Some experiments comparing the performance of the proposed algorithm with a classical word clustering algorithm were carried out.

La Proteómica, otra cara de la genómica

Las proteinas son lo que uno podria llamar los arquitectos de la vida, pues son cruciales en los procesos celulares de todos los seres vivos. Las proteinas estan implicadas en la catalisis de las reacciones quimicas celulares, el transporte de moleculas, la transduccion de senales, la segregacion del material genetico, la produccion y el manejo de la energia. El programa celular vital necesita del trabajo coordinado de muchos tipos diferentes de proteinas (1). La mayor parte del peso seco de una celula esta constituida por proteinas. Parece una tautologia, pero tendremos que entender las proteinas antes de que podamos entender la celula.

Augmented Terrain-Based Navigation to Enable Persistent Autonomy for Underwater Vehicles

To effectively examine ocean processes, sampling campaigns require persistent autonomous underwater vehicles that are able to spend a majority of their deployment time maneuvering and gathering data underwater. Current navigation techniques rely either on high-powered sensors (e.g. Doppler Velocity Loggers) resulting in decreased deployment time, or dead reckoning (compass and IMU) with motion models resulting in poor navigational accuracy due to unbounded sensor drift. Recent work has shown that terrain-based navigation can augment existing navigation methods to bound sensor drift and reduce error in an energy-efficient manner. In this paper, we investigate the augmentation of terrain-based navigation with in situ science data to further increase navigation and localization accuracy. The motivation for this arises from the need for underwater vehicles to navigate within a spatiotemporally dynamic environment and gather data of high scientific value. We investigate a method to create a terrain map with maximum variability across the range of data available. These data combined with local bathymetry create a terrain that enables underwater vehicles to navigate and localize 1) relative to interesting water properties, and 2) globally based on the terrain and traditional methods. We examine a dataset of bathymetry and multiple science parameters gathered at the ocean surface at Big Fishermans Cove on Santa Catalina Island and present a weighting for each parameter. We present efficient algorithms to obtain a convex combination of science and bathymetry parameters for unique trajectories generation.

Minimalist Robot Navigation and Coverage Using a Dynamical System Approach

Equipped only with a clock and a contact sensor, a mobile robot can be programmed to bounce off walls in a predictable way: the robot drives forward until meeting an obstacle, then rotates in place and proceeds forward again. Though this behavior is easily modeled and trivially implemented, is it useful?We present an approach for solving both navigation and coverage problems using such a bouncing robot. The former entails finding a path from one pose to another, while the latter combines different paths over desired locations. Our approach has the following steps: 1) A directed graph is constructed from the environment geometry using the simple bouncing policies, 2) The shortest path on the graph, for navigation, is generated between either one given pair of initial and goal poses or all possible pairs of initial and goal poses, 3) The optimal distribution of bouncing policies is computed so that the actual coverage distribution is as close as possible to the target coverage distribution. Finally, we present experimental results from multiple simulations and hardware experiments to demonstrate the practical utility of our approach.