We review a neuroplanner architecture for use in constructing subcognitive controllers and new ap... more We review a neuroplanner architecture for use in constructing subcognitive controllers and new application that uses it. These controllers have wo important properties: (1) the ability to learn the topology of three continuous spaces: a steering space, a control space, and an observation space, and (2) the ability to integrate the three spaces so that initial and goal steering conditions can suggest a sequence of control states that lead the controlled system to the goal in the presence of obstacles. The result is a rudimentary planner or guidance system that can be used for such subcognitive tasks as robot manipulator control, head/eye coordination, and task sequencing. In this paper, we consider the second domain. The term neuroplanner is intended to convey the impression that the planner is implemented neurally and is more rudimentary than the conventional symbolic planners typical of artificial intelligence research.
The reactive touch model: determining which way to move to avoid touching
1990 IJCNN International Joint Conference on Neural Networks, 1990
A preliminary model is presented that can serve as a starting point for more advanced systems. Th... more A preliminary model is presented that can serve as a starting point for more advanced systems. The model shows how touch sensors can be added to arbitrary points on a robot arm and used for providing reactive information, i.e. information about legal directions for movement when sensors are activated. More specifically, the model can provide detailed information about directions in configuration space that lead to decreasing contact between the arm and obstacles. The model is an extension of the D.H. Graf and W.R. LaLonde (1988) neuroplanar architecture
Proc. of the 1st Int. Workshop. on Agent-Oriented Information Systems, Jun 1, 1999
The recent introduction of many agent-based systems and the increase in agent publications have m... more The recent introduction of many agent-based systems and the increase in agent publications have made the area of agent-based systems a popular area. However, a number of challenges must be met before agent-based system development can claim to be fully mature. One of the most important challenges is the use of a comprehensive design approach whereby the design models fully support agent aspects at the design level. This paper describes a design methodology which allows the development of agent based ...
Technical Correspondence: Comments on Soisalon-Soininen's ``Inessential Error Entries and Their Use in LR Parser Optimization
ACM Transactions on Programming Languages and Systems, 1984
U Recently, Soisalon-Soininen [8] provided a necessary and sufficient condition for error entries... more U Recently, Soisalon-Soininen [8] provided a necessary and sufficient condition for error entries in any deter-ministic LR(1)-based parser to be classified "don't care." This result allows a previous algorithm by Aho and Ullman [2, 3] for partially solving the single-production elimination problem to be used on LALR(1) parsers that employ default reductions. The result is important because it emphasizes the care that must be taken in constructing such optimized parsers; that is, it is easy to produce parsers that lose the ability to detect errors in certain cases (e.g., see [1, 5-7]). Soisa-lon-Soininen solves the problem by eliminating only those single productions for which both default reductions and error detection are not compromised. Although his result allows the elimination of single productions to be maximized, it is unsatisfying because the elimination is not complete. The problem lies with the LALR(1) parsers constructed by extending the Knuth LR(0) tables; essentially, they are too compact. Short of deriving an algorithm that "uncompacts" the tables through some state-splitting scheme, it is not possible to fully eliminate reductions by single productions while still retaining the ability to use default reductions. In order to take full advantage of the Soisalon-Soininen [8] result , it is essential to use variations of the Knuth LR(0) tables that are not "too compacted", for example, Ander-son's [4] LR(0) tables. With a simple grammatical transformation, the standard Knuth algorithm can be used to obtain the Anderson Tables. More important , the Soisalon-Soininen result should be used for achieving a space optimization on the Anderson tables since reductions by single productions have already been eliminated in the latter. Note that such an optimization was still an open problem prior to the Soisalon-Soininen contribution that solved it. In order to produce efficient LALR-based parsers with tables stored in list form, two speed-increasing transformations are highly desirable: (1) removal of reductions by single productions , and (2) use of default reductions. Consider grammar G1 and its corresponding LALR(1) tables 7'1 (Figure 1). If we remove the reductions by single productions using, for example, Aho and Ullman's [1] algorithm, the resulting tables T2 (Figure 2) are obtained. If we modify T1 to have default reductions , the result is T3 (Figure 3). The problem arises when both transformations are applied simultaneously on T~ to get T4 (Figure 4). Using T4, it is easy to see that illegal string i;; is accepted (no error is …
Proceedings. 1987 IEEE International Conference on Robotics and Automation
Programming languages for robotics applications are continually being developed and extended as t... more Programming languages for robotics applications are continually being developed and extended as the applications become more sophisticated. Language evolution is proceeding along two directions: (1) providing more and better facilities for task-level as opposed to robot-level programming and (2) providing better facilities for simulation, graphics and symbolic manipulation. The trend makes it clear that the full capabilities of a general purpose programming language are needed. Instead of developing a new language from the ground up, it is easier and more productive to take an existing language with all the requisite general purpose facilities and specialize it for robotics. Because of its symbolic processing facilities, its object-oriented nature, its usefulness as a simulation language, and its sophisticated graphical interface, Smalltalk is an ideal candidate for specialization. We discuss in more detail why this is the case and we show how a programming language that approaches the power of AL can be imbedded in Smalltalk within 2-4 person-months of effort.
Proceedings of the 1982 SIGPLAN symposium on Compiler construction - SIGPLAN '82, 1982
Traditionally, compilers have been internally organized into separate phases that are invoked in ... more Traditionally, compilers have been internally organized into separate phases that are invoked in a fixed, predetermined order. However, the optimal phase invocation order is usually different for different contexts, such as (I) large batch-oriented production environments, (2) small batch-oriented student environments, and (3) interactive programming environments. We discuss the constraints that must be met in order to obtain a compiler with flexible phase ordering, a property that allows the compiler to be designed and implemented without consideration for the intended operating environment. Flexible phase ordering allows the compiler to be specially configured after the fact. A production compiler based on this principle is also discussed.
ACM Transactions on Programming Languages and Systems, 1981
A simple linear time and space algorithm that transforms a standardized tree representation of an... more A simple linear time and space algorithm that transforms a standardized tree representation of an arithmetic expression lacking operator precedence information into a tree containing it is presented. The approach has applications in modularizing compilers, since the transformations can be applied after the parsing phase. Moreover, it facilitates the compiling of languages that allow dynamic operator definition.
ACM Transactions on Programming Languages and Systems, 1981
Although determinisitic pushdown automata and LR(k) parsers are formally equivalent, the latter (... more Although determinisitic pushdown automata and LR(k) parsers are formally equivalent, the latter (unlike the former) stack one symbol for each input symbol read. By relaxing this requirement, we obtain a class of LR parsers that can perform finite state recognition locally without associated stack manipulations. The definition and construction of such stack-controlling LR parsers is examined in detail along with some potential application areas.
Proceedings of the 3rd ACM SIGACT-SIGPLAN symposium on Principles on programming languages - POPL '76, 1976
A chain production is a production of the form A->M where A is a nonterminal and M is either a te... more A chain production is a production of the form A->M where A is a nonterminal and M is either a terminal or nonterminal. Pager in [PagS] has presented an algorithm which removes all chain reductions from LR(1) parsers after they have been constructed. In this paper, we present an algorithm for directly constructing LR(k) parsers with arbitrary subsets of the chain productions, called the useless chain productions, optimized out. If this subset is empty, the algorithm is a standard one [And i, Kn]. If this subset consists of all chain productions, the result is a parser with all chain reductions optimized away. The algorithm, as in [PagS], also eliminates from the parsers all nonterminals which occur as the left part of useless chain productions. This latter optimization along with the chain reduction optimization significantly decreases the storage space and execution times of the parsers. This provides an efficient solution of the open problem posed by Aho and Ullman [A&U2] for all LR(k) grammars.
Uploads
Papers by Wilf Lalonde