Brian C. Williams

Diagnosing multiple faults

Diagnostic tasks require determining the difierences between a model of an artifact and the artifact itself. The difierences between the manifested behavior of the artifact and the predicted behavior of the model guide the search for the difierences between the artifact and its model. The diagnostic procedure presented in this paper is model-based, inferring the behavior of the composite device from knowledge of the structure and function of the individual components comprising the device. The system (GDE | General Diagnostic Engine) has been implemented and tested on many examples in the domain of troubleshooting digital circuits. This research makes several novel contributions: First, the system diagnoses failures due to multiple faults. Second, failure candidates are represented and manipulated in terms of minimal sets of violated assumptions, resulting in an e‐cient diagnostic procedure. Third, the diagnostic procedure is incremental, exploiting the iterative nature of diagnosis. Fourth, a clear separation is drawn between diagnosis and behavior prediction, resulting in a domain (and inference procedure) independent diagnostic procedure. Fifth, GDE combines modelbased prediction with sequential diagnosis to propose measurements to localize the faults. The normally required conditional probabilities are computed from the structure of the device and models of its components. This capability results from a novel way of incorporating probabilities and information theory into the context mechanism provided by AssumptionBased Truth Maintenance.

Remote Agent: to boldly go where no AI system has gone before

Abstract Renewed motives for space exploration have inspired NASA to work toward the goal of establishing a virtual presence in space, through heterogeneous fleets of robotic explorers. Information technology, and Artificial Intelligence in particular, will play a central role in this endeavor by endowing these explorers with a form of computational intelligence that we call remote agents . In this paper we describe the Remote Agent, a specific autonomous agent architecture based on the principles of model-based programming, on-board deduction and search, and goal-directed closed-loop commanding, that takes a significant step toward enabling this future. This architecture addresses the unique characteristics of the spacecraft domain that require highly reliable autonomous operations over long periods of time with tight deadlines, resource constraints, and concurrent activity among tightly coupled subsystems. The Remote Agent integrates constraintbased temporal planning and scheduling, robust multi-threaded execution, and model-based mode identification and reconfiguration. The demonstration of the integrated system as an on-board controller for Deep Space One, NASAs first New Millennium mission, is scheduled for a period of a week in mid 1999. The development of the Remote Agent also provided the opportunity to reassess some of AIs conventional wisdom about the challenges of implementing embedded systems, tractable reasoning, and knowledge representation. We discuss these issues, and our often contrary experiences, throughout the paper.

Mode Estimation of Probabilistic Hybrid Systems

Model-based diagnosis and mode estimation capabilities excel at diagnosing systems whose symptoms are clearly distinguished from normal behavior. A strength of mode estimation, in particular, is its ability to track a systems discrete dynamics as it moves between different behavioral modes. However, often failures bury their symptoms amongst the signal noise, until their effects become catastrophic.We introduce a hybrid mode estimation system that extracts mode estimates from subtle symptoms. First, we introduce a modeling formalism, called concurrent probabilistic hybrid automata (cPHA), that merge hidden Markov models (HMM) with continuous dynamical system models. Second, we introduce hybrid estimation as a method for tracking and diagnosing cPHA, by unifying traditional continuous state observers with HMM belief update. Finally, we introduce a novel, any-time, any-space algorithm for computing approximate hybrid estimates.

Doing time: putting qualitative reasoning on firmer ground

Recent work in qualitative reasoning has focused on predicting the dynamic behavior of continuous physical systems. Significant headway has been made in identifying the principles necessary to predict this class of behavior. However, the predictive inference engines based on these principles are limited in their ability to reason about time. This paper presents a general approach to behavioral prediction which overcomes many of these limitations. Generality results from a clean separation between principles relating to time, continuity, and qualitative representations. The resulting inference mechanism, based on propagation of constraints, is applicable to a wide class of physical systems exhibiting discrete or continuous behavior, and can be used with a variety of representations (e.g., digital, quantitative, qualitative or symbolic abstractions). In addition, it provides a framework in which to explore a broad range of tasks including prediction, explanation, diagnosis, and design.

A Probabilistic Particle-Control Approximation of Chance-Constrained Stochastic Predictive Control

Robotic systems need to be able to plan control actions that are robust to the inherent uncertainty in the real world. This uncertainty arises due to uncertain state estimation, disturbances, and modeling errors, as well as stochastic mode transitions such as component failures. Chance-constrained control takes into account uncertainty to ensure that the probability of failure, due to collision with obstacles, for example, is below a given threshold. In this paper, we present a novel method for chance-constrained predictive stochastic control of dynamic systems. The method approximates the distribution of the system state using a finite number of particles. By expressing these particles in terms of the control variables, we are able to approximate the original stochastic control problem as a deterministic one; furthermore, the approximation becomes exact as the number of particles tends to infinity. This method applies to arbitrary noise distributions, and for systems with linear or jump Markov linear dynamics, we show that the approximate problem can be solved using efficient mixed-integer linear-programming techniques. We also introduce an important weighting extension that enables the method to deal with low-probability mode transitions such as failures. We demonstrate in simulation that the new method is able to control an aircraft in turbulence and can control a ground vehicle while being robust to brake failures.

Conflict-directed A* and its role in model-based embedded systems

Artificial Intelligence has traditionally used constraint satisfaction and logic to frame a wide range of problems, including planning, diagnosis, cognitive robotics and embedded systems control. However, many decision making problems are now being re-framed as optimization problems, involving a search over a discrete space for the best solution that satisfies a set of constraints. The best methods for finding optimal solutions, such as A^*, explore the space of solutions one state at a time. This paper introduces conflict-directed A^*, a method for solving optimal constraint satisfaction problems. Conflict-directed A^* searches the state space in best first order, but accelerates the search process by eliminating subspaces around each state that are inconsistent. This elimination process builds upon the concepts of conflict and kernel diagnosis used in model-based diagnosis [J. de Kleer, B.C. Williams, Diagnosing multiple faults, Artif. Intell. 32(1) (1987) 97-130; J. de Kleer, A. Mackworth, R. Reiter, Characterizing diagnoses and systems, Artif. Intell. 56 (1992) 197-222] and in dependency-directed search [R. Stallman, G.J. Sussman, Forward reasoning and dependency-directed backtracking in a system for computer-aided circuit analysis, Artif. Intell. 9 (1977) 135-196; J. Gaschnig, Performance measurement and analysis of certain search algorithms, Technical Report CMU-CS-79-124, Carnegie-Mellon University, Pittsburgh, PA, 1979; J. de Kleer, B.C. Williams, Back to backtracking: controlling the ATMS, in: Proceedings of AAAI-86, 1986, pp. 910-917; M. Ginsberg, Dynamic backtracking, J. Artif. Intell. Res. 1 (1993) 25-46]. Conflict-directed A^* is a fundamental tool for building model-based embedded systems, and has been used to solve a range of problems, including fault isolation [J. de Kleer, B.C. Williams, Diagnosing multiple faults, Artif. Intell. 32(1) (1987) 97-130], diagnosis [J. de Kleer, B.C. Williams, Diagnosis with behavioral modes, in: Proceedings of IJCAI-89, 1989, pp. 1324-1330], mode estimation and repair [B.C. Williams, P. Nayak, A model-based approach to reactive self-configuring systems, in: Proceedings of AAAI-96, 1996, pp. 971-978], model-compilation [B.C. Williams, P. Nayak, A reactive planner for a model-based executive, in: Proceedings of IJCAI-97, 1997] and model-based programming [M. Ingham, R. Ragno, B.C. Williams, A reactive model-based programming language for robotic space explorers, in: Proceedings of ISAIRAS-01, 2001].

An autonomous spacecraft agent prototype

This paper describes the New Millennium Remote Agent (NMRA) architecture for autonomous spacecraft control systems. The architecture supports challenging requirements of the autonomous spacecraft domain not usually addressed in mobile robot architectures, including highly reliable autonomous operations over extended time periods in the presence of tight resource constraints, hard deadlines, limited observability, and concurrent activity. A hybrid architecture, NMRA integrates traditional real-time monitoring and control with heterogeneous components for constraint-based planning and scheduling, robust multi-threaded execution, and model-based diagnosis and reconfiguration. Novel features of this integrated architecture include support for robust closed-loop generation and execution of concurrent temporal plans and a hybrid procedural/deductive executive.

Improved human-robot team performance using chaski, a human-inspired plan execution system

We describe the design and evaluation of Chaski, a robot plan execution system that uses insights from human-human teaming to make human-robot teaming more natural and fluid. Chaski is a task-level executive that enables a robot to collaboratively execute a shared plan with a person. The system chooses and schedules the robots actions, adapts to the human partner, and acts to minimize the humans idle time. We evaluate Chaski in human subject experiments in which a person works with a mobile and dexterous robot to collaboratively assemble structures using building blocks. We measure team performance outcomes for robots controlled by Chaski compared to robots that are verbally commanded, step-by-step by the human teammate. We show that Chaski reduces the humans idle time by 85%, a statistically significant difference. This result supports the hypothesis that human-robot team performance is improved when a robot emulates the effective coordination behaviors observed in human teams.

A probabilistic approach to optimal robust path planning with obstacles

Autonomous vehicles need to plan trajectories to a specified goal that avoid obstacles. Previous approaches that used a constrained optimization approach to solve for finite sequences of optimal control inputs have been highly effective. For robust execution, it is essential to take into account the inherent uncertainty in the problem, which arises due to uncertain localization, modeling errors, and disturbances. Prior work has handled the case of deterministically bounded uncertainty. We present here an alternative approach that uses a probabilistic representation of uncertainty, and plans the future probabilistic distribution of the vehicle state so that the probability of collision with obstacles is below a specified threshold. This approach has two main advantages; first, uncertainty is often modeled more naturally using a probabilistic representation (for example in the case of uncertain localization); second, by specifying the probability of successful execution, the desired level of conservatism in the plan can be specified in a meaningful manner. The key idea behind the approach is that the probabilistic obstacle avoidance problem can be expressed as a disjunctive linear program using linear chance constraints. The resulting disjunctive linear program has the same complexity as that corresponding to the deterministic path planning problem with no representation of uncertainty. Hence the resulting problem can be solved using existing, efficient techniques, such that planning with uncertainty requires minimal additional computation. Finally, we present an empirical validation of the new method with a number of aircraft obstacle avoidance scenarios

Scaffolding a Caenorhabditis nematode genome with RNA-seq

Efficient sequencing of animal and plant genomes by next-generation technology should allow many neglected organisms of biological and medical importance to be better understood. As a test case, we have assembled a draft genome of Caenorhabditis sp. 3 PS1010 through a combination of direct sequencing and scaffolding with RNA-seq. We first sequenced genomic DNA and mixed-stage cDNA using paired 75-nt reads from an Illumina GAII. A set of 230 million genomic reads yielded an 80-Mb assembly, with a supercontig N50 of 5.0 kb, covering 90% of 429 kb from previously published genomic contigs. Mixed-stage poly(A)(+) cDNA gave 47.3 million mappable 75-mers (including 5.1 million spliced reads), which separately assembled into 17.8 Mb of cDNA, with an N50 of 1.06 kb. By further scaffolding our genomic supercontigs with cDNA, we increased their N50 to 9.4 kb, nearly double the average gene size in C. elegans. We predicted 22,851 protein-coding genes, and detected expression in 78% of them. Multigenome alignment and data filtering identified 2672 DNA elements conserved between PS1010 and C. elegans that are likely to encode regulatory sequences or previously unknown ncRNAs. Genomic and cDNA sequencing followed by joint assembly is a rapid and useful strategy for biological analysis.