Deepak Khemani

This paper presents use of a cost graph as a representation of a multi-floor building to enable the multi-floor autonomous navigation capability for a team of robot(s). A method for global path planning on this cost graph have been presented. A navigation stack provides a framework for building autonomous navigation capabilities. A navigation stack which enables use of the proposed approach for navigation in a multi-floor building and enables multi-robot operations has been detailed. The improvements provided by the proposed navigation stack over the existing ROS (Robot Operating System) navigation stack have been explained. A way to integrate multiple local path-execution nodes which can combine together to execute the planned global path has also been explained. The paper also demonstrates the reuse of existing ROS compliant source codes for implementation of the proposed navigation stack, thereby optimizing the use of proven and established technology. Further, the extensions to different components of the existing ROS navigation stack, definition of new ROS messages and action definitions, to enable interaction between the components of the stack has been explained. The paper concludes with a brief study on how the proposed stack can be used for multi-robot operations.

Given the diversity and spatio-temporal scales of dynamic coastal processes, sampling is a challenging task for oceanographers. To meet this challenge new robotic platforms such as Autonomous Underwater Vehicle (AUV) are being increasingly used. For effective water sampling during a mission an AUV should be adaptive to its environment, which requires it to be able to identify these dynamic and episodic ocean features in-situ. We describe the use of Hidden Markov Models (HMM) as a feature detection model used onboard an AUV, an autonomous untethered robot. We show how to build an identification model from data collected during past missions. Then we show how the parameters of the HMM can be optimized using a Genetic Algorithm approach, from models trained with the Baum-Welch algorithm in the initial population. Copyright © 2011, Association for the Advancement of Artificial Intelligence. All rights reserved.

Clusters of text documents output by clustering algorithms are often hard to interpret. We describe motivating real-world scenarios that necessitate reconfigurability and high interpretability of clusters and outline the problem of generating clusterings with interpretable and reconfigurable cluster models. We develop a clustering algorithm toward the outlined goal of building interpretable and reconfigurable cluster models; it works by generating rules with disjunctions and conditions on the frequencies of words, to decide on the membership of a document to a cluster. Each cluster is comprised of precisely the set of documents that satisfy the corresponding rule. We show that our approach outperforms the unsupervised decision tree approach by huge margins. We show that the purity and f-measure losses to achieve interpretability are as little as 5% and 3% respectively using our approach. Copyright 2009 ACM.

The quality of a learning algorithm is characterized by the accuracy, stability and comprehensibility of the models it generates. Though ensembles produce accurate and stable classifiers, they are hard to interpret. In this paper, we propose a meta-learning method for ID3 that makes use of an ensemble for gaining accuracy and stability and yet produces a single comprehensible classifier. The main idea here is to generate additional examples at every stage of the decision tree construction process and use them to find the best attribute test. These new examples are classified using the ensemble constructed from the original training set. The number of new examples generated depends on the size of the input attribute space and the input attribute values of new examples are partially determined by the algorithm. Existing work in this area deals with the generation of a fixed number of random examples. Experimental analysis shows that our approach is superior to the existing work in retaining accuracy and stability gains provided by the ensemble classifier. Copyright © 2005, American Association for Artificial Intelligence (www.aaai.org). All rights reserved.

This paper describes a system that uses a hybrid of quantitative and qualitative knowledge to solve physics word problems. Such an integration of knowledge from two models is useful to come up with the correct solution for many of these problems. We have applied this hybrid model to solve word problems from projectile motion. These types of word problems have not been addressed in recent times. We have solved a set of 30 problems in this domain.

The task of planning in a dynamic and unoeitain domain is considerably more challenging than in domains traditionally adopted by classical planning methods. Planning in real situations has to be a knowledge intensive process, particularly since it is not easy to predict all the effects of one's actions. However, many knowledge based implementations are susceptible to brittleness. Contract bridge offen a domain in which many of the issues invohed in real world problems can be addressed without having to make simplifications in representation. Planning in the game of bridge Ukes us away from the traditional search basedmethods(like thealpha.beta procedure), which are applicable in complete infonnation games like chess. In this paper we look at a more flexible use of knowledge structures for planning in bridge.

Recent research in AI Planning is focused on improving the quality of the generated plans. PDDL3 incorporates hard and soft constraints on goals and the plan trajectory. Plan trajectory constraints are conditions that need to be satisfied at various stages of the plan. Soft goals are goals, which need not necessarily be achieved but are desirable. An extension of Constraint Satisfaction Problem, called Optimal Constraint Satisfaction Problem (OCSP) has allowance for defining soft constraints and objective functions. Each soft constraint is associated with a penalty, which will be levied if the constraint is violated. The OCSP solver arrives at a solution that minimizes the total penalty (Objective function) and satisfies all hard constraints. In this paper, an OCSP encoding for the classical planning problems with plan trajectory constraints, soft and hard goals is proposed. Modal operators associated with hard goals and hard plan trajectory constraints are handled by preprocessing and imposing new constraints over the existing GP-CSP encoding. A new encoding for each of the modal operators associated with the soft goals and soft plan trajectory constraints is proposed. Also, a way of encoding conditional goal preference constraints into OCSP is discussed. Based on this research, we intend to submit a planner for the coming planning competition-IPC2006.

The AI community has always been interested in designing intelligent agents which function in a multi-agent arrangement or a man-machine scenario. More often than not, such settings may require agents to work autonomously (or under intermittent supervision at the least) in partially observable environments. Over the last 10 years or so, the planning community has started looking at this interesting class of problems from an epistemic standpoint, by augmenting the notions of knowledge and beliefs to AI planning. In this paper, we present a system that synthesizes plans from the primary agent's perspective, based on its subjective knowledge, in a multi-agent environment. We adopt a semantic approach to represent the mental model of the primary agent whose uncertainty about the world is represented using Kripke's possible worlds interpretation of epistemic logic. Planning in this logical framework is computationally challenging, and, to the best of our knowledge, most of the existing planners work with the notion of knowledge, instead of an agent's subjective knowledge. We demonstrate the system's capability of projecting beliefs of the primary agent on to others, reasoning about the role of other agents in the prospective plans, and preferring the plans that hinge on the primary agent's capabilities to those which demand others' cooperation. We evaluate our system on the problems discussed in the literature and show that it takes fractions of seconds to search for a plan for a given problem. We also discuss the issues that arise in modeling dynamic domains with the representation our system employs.

In this paper, we look into the problem of filtering problem solution repositories (from sources such as community-driven question answering systems) to render them more suitable for usage in knowledge reuse systems. We explore harnessing the fuzzy nature of usability of a solution to a problem, for such compaction. Fuzzy usabilities lead to several challenges; notably, the trade-off between choosing generic or better solutions. We develop an approach that can heed to a user specification of the trade-off between these criteria and introduce several quality measures based on fuzzy usability estimates to ascertain the quality of a problem-solution repository for usage in a Case Based Reasoning system. We establish, through a detailed empirical analysis, that our approach outperforms state-of-the-art approaches on virtually all quality measures. © 2011 ACM.

Plan adaptation is useful in faster generation of plans. We propose a domain independent algorithm for adapting plans with durative actions. Given a plan which is to be adapted to the new problem, our algorithm retains a portion of the old plan (set of actions) that can be useful in solving the new problem. The actions in the portion of the old plan are adjusted to the new durations. These actions are applied one after another according to their start times, starting from the initial state of the new problem. The state with the lowest heuristic value among the intermediate states obtained after application of each action is selected. A plan is found out from this state to the goal state of the problem and added to the set of actions applied to arrive at this state. The resulting set of actions forms the plan for the new problem. Experimental results show that adapting a temporal plan using our algorithm can be very efficient.