Mental State

UploadLog

Thu, 01 Jul 2010 20:11:04 GMT

date	user	location	storeUrl	uploadDir	toFilename	origin
14/06/2010 16:29:02	YourName	scrapbook.html	scrapup.php	./	index.html	ok
14/06/2010 17:10:19	YourName	scrapbook.html	scrapup.php	./	index.html	ok
14/06/2010 20:18:37	pno	scrapbook.html	scrapup.php	.	index.html	ok
14/06/2010 20:37:33	pno	scrapbook.html	scrapup.php	.	index.html	ok
15/06/2010 22:20:50	pno	scrapbook.html	scrapup.php	.	index.html	ok
18/06/2010 14:28:34	pno	scrapbook.html	scrapup.php	.	index.html	ok
18/06/2010 14:34:07	pno	scrapbook.html	scrapup.php	.	index.html	ok
22/06/2010 18:12:51	pno	scrapbook.html	scrapup.php	.	index.html	ok
22/06/2010 18:17:30	pno	scrapbook.html	scrapup.php	.	index.html	ok
01/07/2010 22:11:04	pno	scrapbook.html	scrapup.php	.	index.html

MathTran

Thu, 01 Jul 2010 15:25:50 GMT

Webservice for on-line converting TeX math formulae to images. Could be quite useful...

mathtran link

Das Stinktier Prinzip

Thu, 01 Jul 2010 11:14:26 GMT

Skunk Theory of Hans Joachim Schellnhuber.

link

GT/Adver scenarios in TAF II brainstorming

Tue, 22 Jun 2010 16:17:00 GMT

Starting point

The final aim of the series of 1-3 meetings with BBO & VLY is to reach a state where we agree on a partial scenario within the frame of the TAF II MxN tracking scenario(s) which in ideal case should satisfy the following:

fit the TAF II framework
in the future extensions should fit the TAS II scenario (should have a potential to become the core TAS II scenario)
provide a solid research basis for at least one of the GT/Adver team members (VLY, BBO)

Results of the 2hrs discussion

We went over a series of smaller candidate scenarios. The two emerging topics seem to revolve around patrolling games and search & capture games. It seems that we want to play these games on graphs, subgraphs of the ground street map of the environment. These topics naturally fit the TAS II scenario. Later we tried to push towards brainstorming on applications of these things in TAF II scenarios, i.e., targets vs. UAVs. One of the feelings I have from that meeting is that we are after a kind of symmetric algorithms which would work for both the smart target, as well as for the UAV in the game.

We came up with some ineteresting scenarios leading to an implementation of 1xN, or MxN tracking scenarios. I.e., provided that the targets try to evade, the planes have to seek strategies allowing them to patrol a set of multiple targets which on the other hand assume the best tracking method on the side of the UAVs.

There were two styles of work which I felt emerged from the discussion. Either to go after a general theoretical result with high significance in the community, or to go after an instance of a problem which the community did not see yet. It seems that VLY and BBO prefer the latter.

An interesting discussion emerged around the 1xN tracking scenario, where the UAV tracks several targets and knows about them, while the targets do not know about other tracked entities. In result, the targets operate in a setting where they need to make locally optimal (or similar) decisions, while the UAV can make globally optimal decisions. What are the interesting problems are the relations of the local optima vs. the global optima. This kind of work would go after the high-significance theoretical approach I would prefer, but BBO&VLY do not seem to be fond of this direction.

We were also seeking overlaps between the AgentC project and the TAF II project. No particular notes emerged.

BTW, are there any overlaps with the DeepA, or some other older project? Can we re-use something from the past???

todo: ask VLY about applications of this

Overall, the meeting was a good brainstorming however with little concrete results. We agreed upon a subsequent one on Thursday where we would like to formulate a concrete scenario which we will work on within the context of the TAF II project.

TAF II meeting 20100621

Tue, 22 Jun 2010 15:51:00 GMT

Notes

JVO & TKO are off the project from 28.06.2010 on
the administrative leadership of teh project is handed over to PNO
the project ends in September 2010, where we should put together the final report
the next intermediary report is due at the end of June 2010, i.e., next week

Next actions & todo's on the TAF II project

TAF II intermediary report
- LCH, JVO, TKO: write partial summaries, deadline is 25.06.2010
- PNO: wrap up and complete the report
Finish the scenario D
- DPA is available for 2 weeks in the summer for the project
- content:
  - finish the greedy algorithm (TKO)
  - finish the resource allocation algorithm (JVO)
  - potential fields based algorithm (??? - PNO)
  - clustering based algorithm (??? - PNO)
  - improve the greedy algorithm (??? - TKO)
  - implement some kind of coordination method for the aircrafts (!!! PNO !!!)
GT/Adver@TAF II (VLY+BBO+PNO)
- we would like to have a bonus content involving some game theory in the TAF II project, i.e., the scenario should: i) be interesting by itself, as well as ii) serve as a preparation platform for the TAS project later this year
- can we re-use some software/methods/etc. from BBO & VLY's work in the past? E.g., DeepA project, or the AgentC project?
MASPlan@TAF II (PNO)
- we would like to put into the final report some theoretical text/summary about the multi-agent planning problem. This should serve as a preparation for the TAS project, but since the two are considered a bundle, we can mention it in this project as well
JazzykJVM (??? - PNO)
- what to do with the interpreter? Can we use it for implementation of the people there?
- BTW, this will have some interactions with the UrbanSim project as well.

Notes on efective management

Mon, 21 Jun 2010 21:54:00 GMT

A very interesting text full of useful bits and pieces:

http://danielwilkerson.com/effective-management.html

Bernoulli processes

Fri, 18 Jun 2010 16:10:00 GMT

What I am after in the TL system specifications with failures note seems to be a etude on Bernoulli Processes

Excerpts:

The number of successes in the first n trials, which has a binomial distribution B(n, p)
The number of trials needed to get r successes, which has a negative binomial distribution NB(r, p)
The number of trials needed to get one success, which has a geometric distribution NB(1, p), a special case of the negative binomial distribution

Cf. also geometric distribution

Information & general systems entropy links

Fri, 18 Jun 2010 12:34:00 GMT

http://danielwilkerson.com/entropy.html
http://en.wikipedia.org/wiki/Entropy_%28information_theory%29
http://en.wikipedia.org/wiki/Entropy_in_thermodynamics_and_information_theory
http://www.eolss.net/ebooks/Sample%20Chapters/C02/E6-46-01-04.pdf

http://www.iep.utm.edu/time/
http://books.google.com/books?id=a25hPrUrsSMC&lpg=PA1&ots=5gWtS9-CQQ&dq=temporal%20logics%20entropy&pg=PA2#v=onepage&q=entropy&f=false
Random worlds and maximum entropy

The Real Science Gap

Tue, 15 Jun 2010 20:30:00 GMT

"This article attempts to explain why the US is struggling in its competition with other countries in the realm of scientific advancement. 'It's not insufficient schooling or a shortage of scientists. It's a lack of job opportunities. Americans need the reasonable hope that spending their youth preparing to do science will provide a satisfactory career.' I can hardly believe that somebody actually understands the present situation. It continues, 'The current approach — trying to improve the students or schools — will not produce the desired result, the experts predict, because the forces driving bright young Americans away from technical careers arise elsewhere, in the very structure of the US research establishment. For generations, that establishment served as the world’s nimblest and most productive source of great science and outstanding young scientists. Because of long-ignored internal contradictions, however, the American research enterprise has become so severely dysfunctional that it actively prevents the great majority of the young Americans aspiring to do research from realizing their dreams.'" (taken from Slashdot)

link

Executable Temporal Logic Specification

Tue, 15 Jun 2010 20:28:00 GMT

Baudinet: Temporal logic programming is complete and expressive - link
- temporal logic programming and the TEMPLOG system. Something to check one day...
Fisher, Wooldridge: Executable Temporal Logic for Distributed A.I. - link
- the MetateM stuff

Vereofy model checker

Tue, 15 Jun 2010 20:28:00 GMT

link

Vereofy provides model checking for component-based systems. Vereofy uses two input languages: a scripting language called RSL (Reo Scripting Language), and a guarded command language, called CARML (Constraint Automata Reactive Module Language). RSL and CARML are equally powerful and rely on the same semantic model. However, due to the nature of RSL and CARML, in most cases a hybrid approach, where CARML is used to provide the interface specifications of components and RSL for specifying the network is preferable.

Interval Temporal Logic

Tue, 15 Jun 2010 20:25:00 GMT

There is such an animal. It is quite related to what we wanted with DCTL*, ro to what process logic does. There obviously are tools for system specification with ITL. Among other features it also features the chop operator.

TL system specifications with failures

Tue, 15 Jun 2010 20:20:00 GMT

Introduction

Let \mathcal{A}_\mathit{ag} {}={} \{(a,\phi)\}, where a's are labels and \phi's are propositional formulae, be a set of specifications of actions in a system. Actions can be seen as STRIPS-like specifications which and can be executed anytime, therefore they do not feature preconditions.

Now let's consider a Kripke model M describing a system involving a single agent and an environment. Let also \mathcal{A} denote the set of all actions and events which can happen in M. We can also write that \mathcal{A}=\mathcal{A}_\mathit{ag} \cup \mathcal{A}_\mathit{env}. \mathcal{A}_\mathit{ag} denote agent's actions and \mathcal{A}_\mathit{env} denote events which can occur in the environment.

Let us also consider the DCTL* language for speaking about properties of the system.

Environment dynamics

One of interesting questions which can be formulated is the following

How can we characterize/quantify environment dynamics w.r.t. the given agent system?

In the following I sketch an interesting answer to this question.

Let (a,\phi)\in {}\mathcal{A}_\mathit{ag}{} Let \omega be the minimal number of iterations of an action a the agent must execute in order to ensure that \phi is satisfied in the final possible world. Formally,

\omega {}={} \min \{ n\in{}\mathbb{N} | \forall (a,\phi)\in{}\mathcal{A}_\mathit{ag}, \sigma\in{} M : M,\sigma {} \models{} [a^n]\lozenge\phi\}

If such a minimal n\in{}\mathbb{N} does not exist, we denote \omega{}={}\infty.

Now, if we perceive the system M as a "game" between the environment and the agent where the agent tries to establish some proposition while the environment is trying to do exactly the opposite, the number \omega is negatively proportional to the rate of the environment dynamics. We have the following observations:

if \omega{}={}0, the system seems to be trivial in the sense that in all states all the effects of all actions from \mathcal{A}_\mathit{ag} hold.
if \omega{}={}1, the system is deterministic. I.e., the environment is passive and agent's actions always succeed. In fact, in this system we can always use Dynamic Logic for reasoning about actions, i.e., [a]{} \phi{} . In the language of DCTL*, that amounts to [a]\bigcirc{}\phi;
now the higher \omega, the more dynamic the environment is. We can actually use \omega's to compare the level of dynamics of various environments.
if \omega{}={}\infty, the system is again rather trivial. In fact, it is hopeless because there is no chance the agent can ensure succeed in such a system. The environment is extremely dynamic.

An interesting observation, or rather an intuition is, that the higher the dynamics of the environment, the shorter planning horizon the agent can use. For longer horizons the design becomes ineffective as it can be shown that for longer planning horizons (intentions/commitments), these will always fail.

Actually, we can try to capture the environment dynamics in a dual/complementary way to \omega. Let's define the number \omega^\prime as the longest interval on which the agent can ensure validity of \phi for some a, s.t. (a,\phi){}\in{}\mathcal{A}_\mathit{ag}. Formally,

\omega^\prime{}={}\{ n\in{}\mathbb{N} | \forall (a,\phi)\in{}\mathcal{A}_\mathit{ag},\sigma\in{}M : M,\sigma{}\models{}\square\lozenge{} [a^n]\square{}\phi \}

Now this is a rather strange notion, as it is of little use (from a agent-program-specification point of view) when we are able to say something like this about agent's actions. But if we modify it a bit to

\omega^\prime{}={}\{ n\in{}\mathbb{N} | \forall (a,\phi)\in{}\mathcal{A}_\mathit{env},\sigma\in{}M : M,\sigma{}\models{}\square\lozenge{} [a^n]\square{}\phi \}

suddenly we arrive to some interesting observations.

If \omega is defined (finite), there must be an action (a_1,\phi_1){}\in{} \mathcal{A}_\mathit{ag}{} for which the \omega is the minimal number of steps to ensure its desired effects. Now necessarily, there must exist a complementary environment event (a_2,\phi_2){}\in{}\mathcal{A}_\mathit{env}{} such that \phi_1 \wedge \phi_2 \models{} \bot. Hence,

\omega=\omega^\prime

(plus minus 1?)

Actually, in an environment where the agent does not know the dynamics apriori, it does not really make sense to specify agent's goals as pure maintenance conditions of the pattern \square{} \phi. Such goals can never be satisfied. Rather, these should be reformulated as repair goals of the pattern \square {}(\neg\phi {}\rightarrow{} [a^*]\lozenge\phi).

Note: In practical settings, the environment dynamics could be estimated by sampling. I.e., execution of arbitrary iterations of some "harmless" actions.

Agent specification

Actually the main question I have is

Is there a (fragment of a) logic which captures exactly the specifications of programs which can be constructed as BSM's over a given set of primitive annotated actions?

In other words

What is the set of temporal formulae which can be satisfied as temporally extended goals by a BSM constructed from a given set of annotated primitive actions?

Obviously it is not the full LTL, but rather some kind of weird (incomplete) fragment of it. Some of the hints:

the agent can only achieve positive effects. I.e., unless it has an action which can achieve \neg\phi, the agent can by its capabilities only ensure \phi. So not all LTL formulae can be constructed.
- given a complex goal, we could construct its negation normal form and evaluate the propositional formulae behind temporal modalities. If these involve propositions which do not correspond to effects of some specified action, or include a negation which does not explicitly follow from the agent's actions's effects, we can safely say that the agent cannot by its own capabilities bring about that goal.
the fragment sought after certainly includes all the compositions of the formulae [a^*]\phi (for each action) which correspond to BSM compositions (we can also add a parallel composition?).
I suspect however that there are more of those... I would like to find a complete fragment/logic...

LTL normal form

Tue, 15 Jun 2010 16:09:00 GMT

Negative normal form

Negation appears only in front of propositional literals

The following identities allow to derive a formula in the NNF:

\neg \neg \phi \equiv \phi
\neg \square \phi \equiv \lozenge \neg \phi
\neg \lozenge \phi \equiv \square \neg \phi
\neg (\varphi \mathcal{U} \psi) \equiv (\neg \varphi) \mathcal{R} (\neg \psi)
\neg (\varphi \mathcal{R} \psi) \equiv (\neg \varphi) \mathcal{U} (\neg \psi)

Notes on PDDL resources

Mon, 14 Jun 2010 18:57:00 GMT

PDDL resources

resources page

Bacchus: The Power of Modeling—a Response to pddl2.1

link

interesting note on pgs 1, 2: That dichotomy is between work on
“domain-independent” planning and “control-intensive” planning. Just as work on control-
intensive planning tends to ignore the applicability and power of state-of-the-art search
algorithms for planning, work on domain-independent planning tends to ignore the power to
be gained, and the requirements imposed, by domain modeling.

Fox & Long: PDDL2.1 : An Extension to PDDL for Expressing Temporal Planning Domains

link

they summarize works on temporal planning up to 2003. They mention the following pieces:
- Smith & Weld: Temporal planning with mutual exclusion reasoning, IJCAI 1999
- Bacchus & Kabanza: Planning for Temporally Extended Goals
- Do * Khambampati: Sapa: a domain-independent heuristic metric temporal planner

PDDL 2.1 features some temporal aspects in introducing actions with durations. In fact, they introduce durative actions which only specify conditions at the beginning of the duration interval, at the end of the interval and some invariants for conditions which hold in all states of the duration. However, no other states (combinations, or path specifications) are allowed/made available in the language. In effect, this language is not expressive enough to support complex temporally extended goals.
another interesting feature in PDDL2.1 is introduction of specifications of continuous resources (such as e.g., time?). A typical example would be an action with unspecified duration which however causes a continuous change in a resource - e.g., fuel amount, or temperature.

on pg. 84, there are several notes on related work, i.e., works on planning with temporal features. I collected the following notes:
- TGP (Smith & Weld) planner takes a similar approach as described above. I.e., even though actions have durations, all their effects are realized at the end of the duration interval. In consequence durative actions are treated as black box intervals, i.e., inaccessible to interruptions, or intereference.
- Sapa system (Do & Khambhampati) allows annotations of time points with effects. I.e., during an action execution something can happen in e.g., 2nd, or 3rd, etc. time point. This seems to be a richer framework which could be able to express e.g., \lozenge, or \square operators...

finally the authors discuss also some related issues about reasoning about time. They make some interesting notes on two problems - and I should take a look at these:
1. the frame problem in temporal reasoning frameworks, and
2. the divided instant problem.

Trk: APL track

Mon, 14 Jun 2010 14:27:00 GMT

Agent-Oriented Programming Languages research track

Executable Temporal Logic Specification (Tue 15 Jun 2010 10:28:00 PM CET)
Interval Temporal Logic (Tue 15 Jun 2010 10:25:00 PM CET)
TL system specifications with failures (Tue 15 Jun 2010 10:20:00 PM CET)
LTL normal form (Tue 15 Jun 2010 06:09:00 PM CET)
Research track Agent-Oriented Programming - collection of ideas and inspirations 20100602 (Wed 09 Jun 2010 04:56:00 PM CET)
Planning for Temporally Extended Goals (Wed 09 Jun 2010 04:06:00 PM CET)
On relationship between LTL vs. BSM (Wed 09 Jun 2010 01:43:00 PM CET)
process logic (Wed 09 Jun 2010 01:35:00 PM CET)
Elaboration Tolerance (Tue 08 Jun 2010 04:04:00 PM CET)
Non-monotonic temporal logics (Thu 03 Jun 2010 05:49:00 PM CET)
Active goals in a BSM (Wed 02 Jun 2010 08:42:00 PM CET)
unifying framework for reasoning about goals (Wed 02 Jun 2010 08:21:00 PM CET)
Note on goal types in APLs (Wed 02 Jun 2010 08:19:00 PM CET)
GTD vs. BDI (Wed 02 Jun 2010 08:04:00 PM CET)
NASA's plan for autonomous robots of next-gen (Sun 09 May 2010 12:43:00 AM CET)
JVO/TKO/MPE influences on the BSM planned paper (Mon 25 Jan 2010 01:28:00 PM CET)
C&L influences on the BSM planned paper (Tue 19 Jan 2010 01:39:00 PM CET)
Behavioral Economics Comes to the Rescue of Retirement Savings (Mon 18 Jan 2010 04:17:00 PM CET)
BSM, 2APL, GOAL and the others (Thu 07 Jan 2010 03:07:00 PM CET)
Agent Dynamic Logics (Tue 05 Jan 2010 02:57:00 PM CET)
BSM - programming cognitive agents (Tue 05 Jan 2010 02:57:00 PM CET)
What is Knowledge Representation? (Tue 20 Oct 2009 10:03:00 AM CET)
AgentSpeak meta-interpreter (Tue 20 Oct 2009 10:01:00 AM CET)
BDI & models of parallelism II (Thu 27 Aug 2009 11:43:00 AM CET)
FSM vs. BSM (Fri 12 Jun 2009 03:44:00 PM CET)
Reactivity and agent oriented programming (Fri 12 Jun 2009 03:43:00 PM CET)
Note on UNDERSPECIFICATION in agent-oriented programming (Fri 12 Jun 2009 03:42:00 PM CET)
Robotics and agent oriented programming (Fri 12 Jun 2009 03:41:00 PM CET)
BDI & models of parallelism (Tue 09 Jun 2009 01:56:00 PM CET)
Evaluating different flavours and extensions of BDI (Tue 09 Jun 2009 01:56:00 PM CET)
Commitment Oriented Programming (Fri 05 Jun 2009 12:46:00 AM CET)
Roles (Fri 05 Jun 2009 12:45:00 AM CET)
Birna van Riemsdijk (Tue 02 Jun 2009 10:40:00 AM CET)
Koen Hindriks (Tue 02 Jun 2009 10:40:00 AM CET)
GOAL (Tue 02 Jun 2009 10:38:00 AM CET)

MainMenu

Wed, 09 Jun 2010 14:56:00 GMT

About Events Recent Changes || || TAS APL MASPlan GT/Adver

Research track Agent-Oriented Programming - collection of ideas and inspirations 20100602

Wed, 09 Jun 2010 14:56:00 GMT

Open tasks:

the thesis summary journal paper
- the BSM framework summary
- DCTL* logic/possibly a process logic variant of the framework
- relationship to BDI
- relationship to the original reactive planning
explore the GTD vs. BDI idea
explore the idea Note on goal types in APLs
check the idea on Active goals in a BSM

SideBarOptions

Wed, 09 Jun 2010 14:56:00 GMT

note permaview collapse all close all »

Trk: MASPlan track

Wed, 09 Jun 2010 14:51:00 GMT

Multi-agent planning research track

Vereofy model checker (Tue 15 Jun 2010 10:28:00 PM CET)
Interval Temporal Logic (Tue 15 Jun 2010 10:25:00 PM CET)
TL system specifications with failures (Tue 15 Jun 2010 10:20:00 PM CET)
Notes on PDDL resources (Mon 14 Jun 2010 08:57:00 PM CET)
Planning for Temporally Extended Goals (Wed 09 Jun 2010 04:06:00 PM CET)
Metric Interval Temporal Logic (Wed 09 Jun 2010 01:27:00 PM CET)
Meeting minutes: MASPlan/replanning meeting 20100601 (Tue 01 Jun 2010 08:57:00 PM CET)
Carmel Domshlak & Ronan Brafman: Planning for Loosely Coupled Multi-Agent Systems (Fri 12 Mar 2010 11:17:00 PM CET)
de Weerdt & Clement: tutorial on Multi-agent planning (Fri 12 Mar 2010 10:33:00 PM CET)
desJardins et al.: A Survey of Research in Distributed, Continual Planning (Tue 26 Jan 2010 04:48:00 PM CET)
de Weerdt, ter Mors & Witteveen: Multi-agent Planning: An introduction to planning and coordination (Tue 26 Jan 2010 04:44:00 PM CET)