Poole, D. (1995). Logic programming for robot control. In C.S. Mellish, editor, Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI95) , pages 150-157, San Mateo, CA: Morgan Kaufmann Publishers.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... In robotics, the control part must be integrated with other functions that benefit from the use of a logical formalism, usually classical logic (e.g. modules for action planning and for localization) 19, 35] An explicit reference to logic is also made by the Cognitive Robotics project [23, 27], where the Situation Calculus is adopted as the underlying model and GOLOG, or related logic based components, are proposed at the reasoning and planning level. These proposals generally see logic as confined to the symbolic domain, and no attempt is made to formalize both reasoning and data ....

Poole, D. (1995). Logic programming for robot control. In C.S. Mellish, editor, Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI95) , pages 150-157, San Mateo, CA: Morgan Kaufmann Publishers.

....of sensor data, each requiring explanation. In general, a hypothesised object will only explain a portion of the sensor data of interest. So a complete explanation will comprise a number of hypothesised objects supplemented with a number of noise terms to explain away the rest of the sensor data [Poole, 1995], Shanahan, 1997] To allow for noise terms, the background theory S must be supplemented with some additional formulae. For example, as well as Axiom (1) from Section 5, we might have, Noise(w) FromTo(w,p1,p2) Let the set F of abducibles comprise a set F a of noise terms each with ....

D.Poole, Logic Programming for Robot Control, Proceedings IJCAI 95, pp. 150--157.

.... to illustrate the theoretical and methodological ideas whose promotion is the paper s main purpose, its design does suggest a sharply bipartite architecture which differs markedly from those hinted at in most current work in the Cognitive Robotics vein [Lesprance, et al. 1994] Kowalski, 1995] [Poole, 1995]. The architecture of the Rug Warrior implementation is behaviour based in the purest sense, and yet serves the purpose of building a symbolic model of the world for which a rigorous denotational account can be supplied. In a sense, this architecture accommodates two extremes in robot design ....

D.Poole, Logic Programming for Robot Control, Proceedings IJCAI 95, pp. 150--157.

....rather than interacting with the whole control system at once. The incremental nature of this model enables programmers to build, execute, edit and refine programs in small steps. There are many other control models for programming robots, such as hierarchical[3] and logic based control systems [5]. A possible extension to this work could be investigating the feasibility of these control models as a basis for other visual languages for robot control. It might also be worthwhile to attempt to combine different control models in one visual programming system. 7.1.4 Planning and Problem ....

David Poole ( 1995). Logic Programming for Robot Control". In: Proceedings of 14th International Joint Conference on AI (IJCAI-95).

....instance makes only sense if the door has already been detected by the visual search. The classical approach to incorporate logical dependencies between a number of instances is to assign a symbol to each of the instances and to design a symbolic algorithm which switches the instances on and off (Poole, 1995). The logical dependencies are encoded in the program structure of the symbolic algorithm. The most prominent example of a discrete symbolic algorithm controlling sub behaviors consisting of continuous dynamical systems are the so called Hybrid Systems (Lemmon et al. 1993) Brockett, 1993) For ....

Poole, D. (1995). Logic programming for robot control. In Proceedings of the 14th International Joint Conference on AI (IJCAI-95), Montreal, Canada.

....forwards. The present paper takes a somewhat different approach, which I will now outline. The method for dealing with non determinism adopted here involves the use of determining fluents (see [Shanahan, 1997, Chapter 15] which are related to the noise terms introduced for the same purpose by Poole [1995]. With the method of determining fluents, actions with nondeterministic effects are transformed into actions with deterministic effects. The trick is simply that the outcome of an action with non deterministic effects is made to depend on a fluent, called a determining fluent, which doesn t appear ....

D.Poole, Logic Programming for Robot Control, Proceedings IJCAI 95, pp. 150-157.

....bump switches. The control of the robot via the model of the world it acquires through abduction. Future implementation is expected to adopt a logic programming approach. Existing work in the Cognitive Robotics vein is likely to be influential here [Lesp rance, et al. 1994] Kowalski, 1995] [Poole, 1995]. Acknowledgements The inspiration for Cognitive Robotics comes from Ray Reiter and his colleagues at the University of Toronto. Thanks to Neelakantan Kartha and Rob Miller. The author is an EPSRC Advanced Research Fellow. ....

D.Poole, Logic Programming for Robot Control, Proceedings IJCAI 95, pages 150-157.

....of richer sensor data than that supplied by the Rug Warrior s simple bump switches. The control of the robot via the model of the world it acquires through abduction. Existing work in the Cognitive Robotics vein is likely to be influential here [Lesp rance, et al. 1994] Kowalski, 1995] [Poole, 1995]. Acknowledgements The inspiration for Cognitive Robotics comes from Ray Reiter and his colleagues at the University of Toronto. Thanks to Neelakantan Kartha and Rob Miller. The author is an EPSRC Advanced Research Fellow. ....

D.Poole, Logic Programming for Robot Control, Proceedings IJCAI 95, pages 150-157.

.... illustrate the theoretical and methodological ideas whose promotion is the paper s main purpose, its design does suggest a sharply bipartite architecture which differs markedly from those hinted at in most current work in the Cognitive Robotics vein [Lesp rance, et al. 1994] Kowalski, 1995] [Poole, 1995]. The architecture of the Rug Warrior implementation is behaviour based in the purest sense, and yet serves the purpose of building a symbolic model of the world for which a rigorous denotational account can be supplied. In a sense, this architecture accommodates two extremes in robot design ....

D.Poole, Logic Programming for Robot Control, Proceedings IJCAI 95, pages 150-157.

....the strategy of reducing goals to abducible sub goals is also employed. Planning by abduction is an increasingly popular technique in logic based systems. It has been used in Allen s temporal logic [All91] and several times in logic programming systems (see [Esh88b] Sha89] Eva89] MBD95] and [Poo95] Most of these systems use the Event Calculus or some event based formalism to deal with the temporal reasoning required by the planning problem. According to Reiter [Rei96] there is no alternative but to use an abductive account of reasoning when an event based temporal logic is involved. ....

....all these components. The system is inspired by the Belief Desire Intention paradigm [Bra87] and was used to control FLAKEY, a robot in a space station scenario. Other systems include IPEM [AIS88] which integrated partial planning with execution and Poole s logic programs for Robot Control [Poo95] All these systems rely on some kind of condition action rules by which inputs are related to outputs. 6.2.2 Criticism of Reactive Planning Reactive planning has been analysed by Ginsberg in [Gin89] who calls it universal planning) He proves that even if the compile time costs of the ....

David Poole. Logic programming for robot control. In Chris S. Mellish, editor, Proc. International Joint Conference on Artificial Intelligence, pages 150--157, San Mateo, California, 1995. Morgan Kaufmann Publishers, Inc.

....systems [29; 13] is common in many areas of science, from mechanical engineering to economics to ecology. We assume a time structure o , that is totally ordered and has a metric over intervals. o can either be continuous or discrete; for this paper we will consider discrete time. See [39] for a development of continuous time in this framework. A trace is a function from into some domain . A transduction is a function from (input) traces into (output) traces that is causal in the sense that the output at time can only depend in inputs at times where (i.e. the ....

....and set of assignments of agents to choices will make sense. Agents have input and outputs; there are some values that they have no access to, and some internal values that only they can access. We will model agents as a logic program that specifies how the outputs are entailed by certain inputs [39] . This logic program can use the internal values and sense values but can t use those values the agent has no access to (i.e. can t sense or otherwise determine) In modelling agents we have to be careful about a number of things: 13 With deterministic agents, only the input history is needed. ....

D. Poole. Logic programming for robot control. In Proc. 14th International Joint Conf. on Artificial Intelligence (IJCAI-95), pages 150--157. 1995.

....partially observable MDPs (POMDPs) where a policy is a function from belief states into actions, it is often more convenient to use a policy tree [Kaelbling et al. 1996] which is much more like a robot plan as developed here see Section 7. Rather than assuming robots have policies [Poole, 1995c] we can instead consider robot plans as in GOLOG [Levesque et al. 1996] These plans consider sequences of steps, with conditions and loops, rather than reactive strategies. In this paper we restrict ourselves to conditional plans; we do not consider loops or nondeterministic choice, although ....

D. Poole. Logic programming for robot control. In Proc. 14th International Joint Conf. on Artificial Intelligence, pages 150--157, ftp://ftp.cs.ubc.ca/ftp/ local/poole/papers/lprc.ps.gz, 1995.

....on the output of potentially noisy sensors) Iteration and local variables are explored briefly in Section 2.11. In other work, we have considered policies within the ICL including multiple agents and noisy sensors (Poole 1997a) We have also investigated continuous time in the same framework (Poole 1995). 1.6 Action Preconditions It may seem as though the definition of an action is uncontroversial. This is perhaps because most formal theories treat actions as primitive, and assume that the available actions are given. However, when we try to build a robot, we soon find out that there isn t a ....

....see Poole (1997a) 25 If the effects aren t independent, we can t do better than considering the combinatorial effects, as the number of independent parameters that needs to be specified is exponential in the number of state variables. 41 for details. We only consider discrete time here. See Poole (1995) for a way to handle continuous time (allowing for integration and differentiation with respect to time) using a method similar to the event calculus. The idea is to represent agents and nature in the same way. For the situation calculus axiomatization above, the single agent was treated quite ....

[Article contains additional citation context not shown here]

Poole, D. (1995). Logic programming for robot control, Proc. 14th International Joint Conf. on Artificial Intelligence (IJCAI-95), Montr'eal, Qu'ebec, pp. 150--157.

....dynamical systems [29; 13] is common in many areas of science, from mechanical engineering to economics to ecology. We assume a time structure T , that is totally ordered and has a metric over intervals. T can either be continuous or discrete; for this paper we will consider discrete time. See [39] for a development of continuous time in this framework. A trace is a function from T into some domain A. A transduction is a function from (input) traces into (output) traces that is causal in the sense that the output at time t can only depend in inputs at times t 0 where t 0 t (i.e. ....

....and set of assignments of agents to choices will make sense. Agents have input and outputs; there are some values that they have no access to, and some internal values that only they can access. We will model agents as a logic program that specifies how the outputs are entailed by certain inputs [39] . This logic program can use the internal values and sense values but can t use those values the agent has no access to (i.e. can t sense or otherwise determine) In modelling agents we have to be careful about a number of things: 13 With deterministic agents, only the input history is needed. ....

D. Poole. Logic programming for robot control. In Proc. 14th International Joint Conf. on Artificial Intelligence (IJCAI-95), pages 150--157. 1995.

....solving partially observable MDPs (POMDPs) where a policy is a function from belief states into actions, it is often more convenient to use a policy tree [Kaelbling et al. 1996] which is much more like a robot plan as developed here see Section 7. Rather than assuming robots have policies [Poole, 1995c] we can instead consider robot plans as in GOLOG [Levesque et al. 1996] These plans consider sequences of steps, with conditions and loops, rather than reactive strategies. In this paper we restrict ourselves to conditional plans; we do not consider loops or nondeterministic choice, although ....

D. Poole. Logic programming for robot control. In Proc. 14th InternationalJoint Conf. on Artificial Intelligence, pages 150--157, ftp://ftp.cs.ubc.ca/ftp/ local/poole/papers/lprc.ps.gz, 1995.

....on the output of potentially noisy sensors) Iteration and local variables are explored briefly in Section 2.10. In other work, we have considered the policies within the ICL including multiple agents and noisy sensors (Poole 1997) We have also investigated continuous time in the same framework (Poole 1995). 1.6 The Situation Calculus and the ICL The independent choice logic (Poole 1997) an extension of probabilistic Horn abduction (Poole 1993) to include multiple agents and negation as failure) is a simple framework consisting of independent choices make by nature (and potentially other agents) ....

....state. We can also incorporate non deterministic actions. 3.3 Independent Choice Logic and Reactive Policies There is a conceptually different way to use the ICL to model time and action. Here we can only sketch the idea; see Poole (1997) for details. We only consider discrete time here. See Poole (1995) for a way to handle continuous time (allowing for integration and differentiation with respect to time) using a method similar to the event calculus. The idea is to represent agents and nature in the same way. For the situation calculus axiomatization above, the single agent was treated quite ....

[Article contains additional citation context not shown here]

Poole, D. (1995). Logic programming for robot control, Proc. 14th International Joint Conf. on Artificial Intelligence (IJCAI-95), Montreal, Quebec, pp. 150--157.

No context found.

D. Poole. Logic programming for robot control. Technical Report, Department of Computer Science, UBC, ftp://ftp.cs.ubc.ca/ftp/local/poole/ papers/lprc.ps.gz, 1995.

No context found.

D. Poole. Logic programming for robot control. In Proc. IJCAI '95, 150--157.

No context found.

D. Poole. Logic programming for robot control. In Proc. IJCAI '95, 150--157.

No context found.

D Poole. Logic programming for robot control. In Proceedings of the 14th International Joint Conference on AI (IJCAI-95), Montreal, Canada, 1995. 114

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC