Chitta Baral and Son Cao Tran. Relating Theories of Actions and Reactive Robot Control. In Proceedings of the AAAI 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... to artificial intelligence often feature situatedness of agents reacting to changes in environment instead of reliance on abstract representation and inferen tial reasoning [2, 5, 15] Under certain assumptions, situated behaviours can be proven to be correct with respect to a theory of actions [1], and in some cases can be derived from it [9] The principal aim of this paper is to relate declarative agent specifications and situated behaviours in the context of multi agent teamwork, where tactical reasoning is required. In other words, we will illustrate how an underlying abstract ....

....express each instantiation (e, s) of the tropistic behaviour function e tropistic behaviour(s) in terms of complete sensory states. Instead, we represent such behaviour instantiations in terms of partial sensory states. For example, the following rule, given in the form similar to control rules [1] or condition action pairs [5] describes behaviour instantiations: if [SeeBall : distance, direction) A Far(distance) then turn(direction) dash( 2distance) The bracketed component on the left hand side corresponds to elements of S and has to be evaluated as true in order to activate ....

Chitta Baral and Son Cao Tran. Relating Theories of Actions and Reactive Robot Control. In Proceedings of the AAAI 1996.

....that reactive behaviours can be derived from (and importantly, can be proved to be correct with respect to) a higher level theory is not new. For instance, connections between theories of actions and reactive robot control architectures based on the paradigm of situated activity are explored in [1, 11]. The approach described in [1] formalises further the concept of an action leading to a goal de ned at the representation level in the situated automata approach [5] and follows the latter in relating declarative agent speci cations and situated behaviours. The view taken in this paper is ....

....derived from (and importantly, can be proved to be correct with respect to) a higher level theory is not new. For instance, connections between theories of actions and reactive robot control architectures based on the paradigm of situated activity are explored in [1, 11] The approach described in [1] formalises further the concept of an action leading to a goal de ned at the representation level in the situated automata approach [5] and follows the latter in relating declarative agent speci cations and situated behaviours. The view taken in this paper is that, rather than de ning situated ....

Chitta Baral and Son Cao Tran. Relating Theories of Actions and Reactive Robot Control. In Proceedings of the AAAI

....that reactive behaviours can be derived from (and importantly, can be proved to be correct with respect to) a higher level theory is not new. For instance, connections between theories of actions and reactive robot control architectures based on the paradigm of situated activity are explored in [1, 10]. The approach described in [1] formalises further the concept of an action leading to a goal defined at the representation level in the situated automata approach [5] and follows the latter in relating declarative agent specifications and situated behaviours. The view taken in this paper is ....

....derived from (and importantly, can be proved to be correct with respect to) a higher level theory is not new. For instance, connections between theories of actions and reactive robot control architectures based on the paradigm of situated activity are explored in [1, 10] The approach described in [1] formalises further the concept of an action leading to a goal defined at the representation level in the situated automata approach [5] and follows the latter in relating declarative agent specifications and situated behaviours. The view taken in this paper is that, rather than defining ....

Chitta Baral and Son Cao Tran. Relating Theories of Actions and Reactive Robot Control. In Proceedings of the AAAI 1996 Workshop on Theories of Action and Planning: Bridging the Gap. Portland 1996.

....reactive behaviours can be proved to be correct with respect to a theory of actions (and in some cases can be derived from it) is relatively new. For instance, connection between theories of actions and reactive robot control architectures based on the paradigm of situated activity is explored in [1]. The approach described in [1] formalises further the concept of an action leading to a goal defined at the representation level in the situated automata approach [12] and follows the latter in relating declarative agent specifications and situated behaviours. Development of agent architectures ....

....to be correct with respect to a theory of actions (and in some cases can be derived from it) is relatively new. For instance, connection between theories of actions and reactive robot control architectures based on the paradigm of situated activity is explored in [1] The approach described in [1] formalises further the concept of an action leading to a goal defined at the representation level in the situated automata approach [12] and follows the latter in relating declarative agent specifications and situated behaviours. Development of agent architectures with a formal model in logic ....

[Article contains additional citation context not shown here]

Baral, C., and Son, T. Relating Theories of Actions and Reactive Robot Control. In Proceedings of the AAAI 1996 Workshop on Theories of Action and Planning: Bridging the Gap. Portland (1996).

....about the environment may not be available and not all the possible execution scenarios are known. To define the execution of workflow agents, we rely on a control module of reactive rules. Such control modules, similar to production rule systems, have been used for real time robot control [10, 42]. A control module is a collection of rules of the form: if e 1 ; e n ; f 1 ; f k ; unk f k 1 ; unk fm then a where: ffl each e i is an event literal (possibly negated) ffl each f i (1 i k) is a (positive or negated) fact ffl each unk f j specifies a particular fact ....

C. Baral and T. Son. Relating theories of actions and reactive robot control. In AAAI 96 Workshop on Reasoning about actions, planning and robot control: bridging the gap, 1996.

....guidelines on how to construct correct triggers from given evolution specifications. Our approach in this paper has some relationship with research in specification and synthesis of concurrent programs [14, 15] and the formalization of the relation between reactive control and theories of actions [5] and automatic construction of reactive control from theories of action [5] In [14, 15] concurrent programs are specified using temporal formulas. They do not use aggregates. In general program synthesis is harder because the allowable statements in a program are more general than allowable ....

.... Our approach in this paper has some relationship with research in specification and synthesis of concurrent programs [14, 15] and the formalization of the relation between reactive control and theories of actions [5] and automatic construction of reactive control from theories of action [5]. In [14, 15] concurrent programs are specified using temporal formulas. They do not use aggregates. In general program synthesis is harder because the allowable statements in a program are more general than allowable modifications to a database. On the other hand the approach in [5] although ....

[Article contains additional citation context not shown here]

C. Baral and T. Son. Relating theories of actions and reactive robot control. In Proc. of AAAI 96 Workshop on Reasoning about actions, planning and robot control: bridging the gap., 1996.

....R 7 R 1 R 2 R 9 R 8 R 3 R 5 R 4 C 6 C 11 C 17 C 12 C 13 C 14 C 15 C 16 C 10 F 1 Foyer Figure 1: A Diagram Showing the Topological Map for Event 1 Participants at the AAAI 96 Robotics Competition. 2 Our theory of agents : a brief description Our theory of agents is detailed in our earlier work [BS96, BS97a, BS97b] In this theory we are concerned with the control of one agent in a dynamic environment, where exogenous events beyond the control of our agent may occur. Our agent has an action theory (such as a theory in the language [GL93, KL94, San94, BGP96] where actions and their effects ....

....of both. All this is supposed to happen off line. Unlike, universal plans [Sch87] we only consider a subset of all states, which are carefully constructed from a given set of plausible states based on what actions the agent is going to do and what exogenous events are highly likely. In [BS96] these set of states is referred to as the closure. Our agent is supposed to be equipped with such a control module before it is on its own in the environment. Once the agent is on line, it is supposed to continuously sense and react based on its control module. In rare cases the agent may be ....

C. Baral and T. Son. Relating theories of actions and reactive robot control. In Proc. of AAAI 96 Workshop on Reasoning about actions, planning and robot control: bridging the gap., 1996.

....D, and a goal G if, S = Closure(S; P; A) and for all states s in S, for every correct unfolding function U of P , and for all f in G, D j= f after UnfoldU (P; s; at s. 2 6 Final remarks In this paper we extended our earlier work on formalizing reactive control using action theories [BS96] to cooperative agents in a workflow setting. We have implemented the workflow described in this paper 8 and it is undergoing extensive testing before it will be used at UTEP. Although there has been a lot of work on multi agent systems ( Sin94] and papers in earlier ATAL workshops) the ....

C. Baral and T. Son. Relating theories of actions and reactive robot control. In Proc. of AAAI 96 Workshop on Reasoning about actions, planning and robot control: bridging the gap., 1996. 8 The interface of our system can be seen at http://cs.utep.edu/chitta/workflow.

....R 7 R 1 R 2 R 9 R 8 R 3 R 5 R 4 C 6 C 11 C 17 C 12 C 13 C 14 C 15 C 16 C 10 F 1 Foyer Figure 1: A Diagram Showing the Topological Map for Event 1 Participants at the AAAI 96 Robotics Competition. 2 Our theory of agents : a brief description Our theory of agents is detailed in our earlier work [BS96, BS97a, BS97b] In this theory we are concerned with the control of one agent in a dynamic environment, where exogenous events beyond the control of our agent may occur. Our agent has an action theory (such as a theory in the language [GL93, KL94, San94, BGP96] where actions and their effects ....

....of both. All this is supposed to happen offline. Unlike, universal plans [Sch87] we only consider a subset of all states, which are carefully constructed from a given set of plausible states based on what actions the agent is going to do and what exogenous events are highly likely. In [BS96] these set of states is referred to as the closure. Our agent is supposed to be equipped with such a control module before it is on its own in the environment. Once the agent is on line, it is supposed to continuously sense and react based on its control module. In rare cases the agent may be ....

C. Baral and T. Son. Relating theories of actions and reactive robot control. In Proc. of AAAI 96 Workshop on Reasoning about actions, planning and robot control: bridging the gap., 1996.

....express or define the control program of agents that allow these flexibilities we will rely on the continuous execution of a control module of condition action rules that are triggered by changes in the workflow environment. Such control modules have been used for real time robot control [Nil94, BS96, Bro86, KR91] They are similar to production systems and early active rules that only had the CA part. A simplified version of a workflow control module corresponding to the registration example is the following: M 1 A sequential control module for student registration REPEAT if ....

....M;A) Let M be a sequential workflow control module and s be the state corresponding to the initial input given to the workflow. Unfold(M; s) is the workflow program obtained by unfolding the control module M with 12 We can formally prove that such a set exists following the ideas in [BS96] respect to s, by assuming that no activities other than the ones dictated by M happen. Note that the occurrences of any other activity will be captured by the Closure definition. For example, the conditional program in Example 1 is obtained by unfolding the control module M 1 with respect ....

[Article contains additional citation context not shown here]

C. Baral and T. Son. Relating theories of actions and reactive robot control. In Proc. of AAAI 96 Workshop on Reasoning about actions, planning and robot control: bridging the gap., 1996. 15 The interface of our system can be seen at http://cs.utep.edu/chitta/workflow.

No context found.

Baral, C., and Son, T. 1996. Relating Theories of Actions and Reactive Robot Control. In Proceedings of the AAAI 1996 Workshop on Theories of Action and Planning: Bridging the Gap. Portland.

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