For an autonomous robot or software agent to participate in the social life of humans, it must have a way to perform a calculus of social behavior, an operational model which would allow it to reason about the implications of its own and the humans' actions. To be useful, such a calculus must satisfy several requirements. First, it must have explanatory power (it must provide a coherent theory for why the humans act the way they do), and predictive power (it must provide some plausible scenarios about the future actions of the humans). The social calculus must also be culture-specific, it needs to consider the different social norms and requirements in various societies. As many social interactions take place in public, it also needs to have a model of public perception. The outcome of the project was the set of social calculus techniques designed to satisfy these requirements. For development of the system we had used Java based agent system YAES and for visualization we incorporated it with Java based virtual world Openwonderland.Download: [AAMAS-11] [FLAIRS-25] [BRIMS-21] [MABS-13] [FLAIRS-26] [HAIDM-13] [AAMAS-13]
Moving in dense crowds requires a balance of assertiveness and politeness. Every culture penalizes people who violate other people's personal space or cut off their movement. Yet, in chaotically moving dense crowds, forward movement is impossible without a credible threat of personal space violation. We model crowd movement as a series of micro-conflicts where participants must decide on the adjustments they must make to their movement patterns to avoid collision or violation of personal space. The micro-conflicts are resolved through a series of two-player games where the player either gives way or continues to move forward. We collect data from human players navigating a crowd in a simulated marketplace and use techniques of imitation learning to teach a robot strategies to behave in micro-conflicts similarly to the human models. We develop the game theoretic framework using Java based agent system YAES. For machine learning we use support vector machines (LibSVM) and random forests implemented in Weka. Further for comparative analysis use ANJI for neuroevolution of augmented topologies (NEAT).Download: [IRS-13] [PAIR-13] [AAMAS-13]
The focus of the project is the catastrophic events that decrease the network's QoS by disconnecting a part of the fully functional nodes from the rest of the network. Hence bridge nodes are formed by the network nodes in the vicinity of the catastrophic region that remained safe during the catastrophe. The bridge node provide a relay connection between sub-networks but are prone to natural faliures too. These failures occur mostly due to the energy depletion in the bridge nodes. The energy is rapidly depleted in the bridge nodes due to the overwhelming data traffic which is relayed through these nodes. We extend the concept of bridge protection algorithm (BPA) which complements the algorithms for federated sensor network. We propose the use of relative neighborhood graph for protecting the bridge nodes energy consumption. Our proposed scheme helps the formation of multipath, non-overlapping paths to increase the network's lifetime.
Body and home area networks are an embodiment of their steady infiltration in our daily lives. Inherent energy issues with the sensor network technology paradigm have brought an emphasis towards the energy efficiency issue, thereby, propelling researchers to focus on developing topologies with minimal energy consumption constraints. The project focused on the development of an efficient commercializable home area networking using Zigbee. We had implemented the self-organizing clustered topology with a low latency periodic and query-based data collection solution and have proved our proposition to have a significant reduction in terms of energy expenditures.Download: [ICET'08] [SAS'09] [SENSORCOMM'09] [ICEE'09]
We developed an agent-based system of a fictional (but feasible) information trading business. We were interested in the general properties of the emerging information market: the amount of realizable profit and its distribution between the trader and customers, the business strategies necessary to keep the market operational (such as promotional deals), the price elasticity of demand and the impact of pricing strategies on the profit. The outcome of the project was The Gas Price Information Trader (GPIT): buys information about real-time gas prices in a metropolitan area from drivers and resells the information to drivers who need to refuel their vehicles. We use real world statistics of gas price fluctuation to create scenarios of temporal and spatial distribution of gas prices. The price of the information is determined on a case-by-case basis through a simple negotiation model. The trader and the customers are adapting their negotiation strategies based on their historical profits. For development of the system we had used Java based agent system YAES.Download: [ISCIS-26]
This project was designed to investigate the fault tolerance capabilities of neural networks developed using evolutionary algorithms. Neuro Evolution of Augmenting Topologies NEAT is well known in the field of neuro-evolution and was used in this work. For implementation of the evolutionary system we had used ANJI which is a Java based NEAT simulator. For the robotics part, we integrated SIMBAD, a java-based 3D robotic environment, with ANJI. We investigated whether fault tolerance capabilities can be developed by adopting special methods in evaluation (training) or through the fitness function. The platform chosen for these experiments required finding a successful controller for a rescue robot operating in harsh operating conditions modeled by simulating failure of sensory inputs. The outcome of project was to achieve graceful performance degradation in the presence of faults.
Last Updated 12-31-2013 by Saad Ahmad Khan, University of Central Florida.