Tod S. Levitt

Qualitative navigation for mobile robots

We have developed a multi-level theory of spatial representation of the environment based upon the observation and re-acquisition of distinctive visual events, i.e., landmarks. The representation provides the theoretical foundations for a visual memory database that includes coordinate-free, topological representation of relative spatial location, yet smoothly integrates available metric knowledge of relative or absolute angles and distances

Model-Based Influence Diagrams for Machine Vision

Abstract We show an approach to automated control of machine vision systems based on incremental creation and evaluation of a particular family of influence diagrams that represent hypotheses of imagery interpretation and possible subsequent processing decisions. In our approach, model-based machine vision techniques are integrated with hierarchical Bayesian inference to provide a framework for representing and matching instances of objects and relationships in imagery and for accruing probabilities to rank order conflicting scene interpretations. We extend a result of Tatman and Shachter to show that the sequence of processing decisions derived from evaluating the diagrams at each stage is the same as the sequence that would have been derived by evaluating the final influence diagram that contains all random variables created during the run of the vision system.

Utility-based control for computer vision

Publisher Summary Several key issues arise in implementing computer vision recognition of world objects in terms of Bayesian networks. Computational efficiency is a driving force. Perceptual networks are very deep, typically fifteen levels of structure. The network is not fixed but is created incrementally at runtime. The generation of hypotheses of world objects and indexing of models for recognition are important. This chapter discusses the near-term implementation with parallel computation in a radar surveillance system, ADRIES, and a system for industrial part recognition, SUCCESSOR. For many applications, vision must be faster to be practical and, therefore, efficiently controlling the machine vision process is critical. Perceptual operators may scan megapixels and may require minutes of computation time. It is necessary to avoid unnecessary sensor actions and computation. Parallel computation is available at several levels of processor capability. The potential for parallel, distributed computation for high-level vision means distributing nonhomogeneous computations. The chapter discusses the problem of task control in machine vision systems based on Bayesian probability models. Maximizing utility allows adopting perceptual strategies for efficient information gathering with sensors and analysis of sensor data. The chapter presents the results of controlling machine vision via utility to recognize military situations.

Terrain models for an autonomous land vehicle

We present an architecture for terrain recognition for an autonomous land vehicle. Basic components of this are a set of data bases for generic object models, perceptual structures, temporary memory for the instantiation of object and relational hypothesis, and a long term memory for storing stable hypothesis which are affixed to the terrain representation. Different inference processes operate over these data bases. We describe components of this architecture: the perceptual structure data base, the grouping processes that operate over this, and schemas. We conclude with a processing example for matching predictions from the long term terrain model to imagery and extracting significant perceptual structures for consideration as potential landmarks.

Computational Inference for Evidential Reasoning in Support of Judicial Proof

The process of judicial proof accrues evidence to confirm or deny hypotheses about world events relevant to a legal case. Software applications that seek to support this process must provide the user with sophisticated capabilities to manipulate evidential reasoning for legal cases. This requires computational techniques to represent the actors, entities, events, and context of world situations to structure alternative hypotheses interpreting evidence and to execute processes that draw inferences about the truth of hypotheses by assessing the relevance and weight of evidence to confirm or deny the hypotheses. Bayesian inference networks are combined with knowledge representations from artificial intelligence to structure and analyze evidential argumentation. The infamous 1994 Raddad murder trial in Nice, France provides a backdrop against which we illustrate the application of these techniques to evidential reasoning in support of judicial proof.

Evidential reasoning for object recognition

The authors present a framework to guide development of evidential reasoning in object recognition systems. Principles of evidential reasoning processes for open-world object recognition are proposed and applied to build evidential reasoning capabilities. The principles summarize research and findings by the authors up through the mid-1990s, including seminal results in object-centered computer vision, figure-ground discrimination, and the application of hierarchical Bayesian inference, Bayesian networks, and decision graphs to evidential reasoning for object recognition.

Bayesian inference for radar imagery based surveillance

Publisher Summary This chapter discusses Bayesian inference for radar imagery based surveillance. Inference is performed over a space of hierarchically linked hypotheses. The hypotheses represent statements of the form there is a military force of type F in deployment D at world location L at time T . The hierarchy in the hypothesis space corresponds to the hierarchy inherent in military doctrine of force structuring. Thus, array-level hypotheses of military units, such as companies, artillery batteries, and missile sites, are linked to their component unit hypotheses of vehicles batteries and missile launchers. Similarly, companies are grouped to form battalion hypotheses and battalions to form regiments. The structure of inference follows a pattern based on the models that are matched to generate hypotheses of the presence of military forces on the battlefield. The chapter discusses a general problem of symbolic hierarchical Bayesian inference for military force inference. It presents a method for approximate hierarchical accrual that can be used to selectively avoid unnecessary conflict resolution depending on the systems focus of attention in processing tasks.

Model-Based Probabilistic Situation Inference in Hierarchical Hypothesis Spaces

Publisher Summary This chapter discusses the model-based probabilistic situation inference in hierarchical hypothesis spaces. Artificial intelligence applications such as industrial robotics, military surveillance, and hazardous environment clean-up require situation understanding based on partial, uncertain, and ambiguous or erroneous evidence. It is necessary to evaluate the relative likelihood of multiple possible hypotheses of the (current) situation faced by the decision-making program. The objective of the hierarchical inference process is to produce a global scene interpretation composed of some consistent set of hypotheses. In general, one has no prior model of the global situation. Even if there is a numerical scheme for accruing the weight of evidence supporting any given object (model) hypothesis, there is still no obvious method for deciding which set of hypothesized objects are the best global interpretation of the scene. The methodology accounts for partial pattern matching in model-driven hypothesis generation. If a best hierarchical interpretation is inferred at the top level, then the hierarchical hypothesis tree associated to it will be optimal at all lower hierarchy levels with probability close to one.

Incremental dynamic construction of layered polytree networks

Certain classes of problems, including perceptual data understanding, robotics, discovery, and learning, can be represented as incremental, dynamically constructed belief networks. These automatically constructed networks can be dynamically extended and modified as evidence of new individuals becomes available. The main result of this paper is the incremental extension of the singly connected polytree network in such a way that the network retains its singly connected polytree structure after the changes. The algorithm is deterministic and is guaranteed to have a complexity of single node addition that is at most of order proportional to the number of nodes (or size) of the network. Additional speed-up can be achieved by maintaining the path information. Despite its incremental and dynamic nature, the algorithm can also be used for probabilistic inference in belief networks in a fashion similar to other exact inference algorithms.

Artificial Intelligence: Uncertainty

Reasoning under uncertainty is a central challenge in designing artificial intelligence (AI) software systems. Sources of uncertainty include equally plausible alternative explanations, missing information, incorrect object and event typing, diffuse evidence, ambiguous references, prediction of future events, and deliberate deception. We provide a brief overview of the history of uncertainty management in AI systems, including perceptrons, first order predicate logics, Bayesian networks, neural networks, fuzzy logic, and the Dempster-Shafer theory. The flavor of the research and corresponding technological capabilities in automated inductive reasoning is illustrated. A limited instance of everyday human cognition is represented using a hybrid symbolic and probabilistic representation along with powerful automated inference solution algorithms. It is concluded that uncertainty management-driven advances in automated inductive reasoning raise the hope that AI systems might soon accelerate our rate of scientific discovery.