Programming today is a race between software engineers striving to build bigger and better idiot-proof programs, and the Universe trying to produce bigger and better idiots. So far, the Universe is winning.
Web Planner is an online interface with an underlying PDDL-based automated planner for deterministic planning. This interface is the result of work from my PhD student Mauricio and an undergraduate student and allows end users to visualise both the formalisation of the operators and the way heuristics evaluate the state space towards the goal. Work on the operator visualisation language was published in the paper DOVETAIL - An abstraction for Classical Planning using a Visual Metaphor, whereas work on the heuristic visualisation was published in the paper WEB PLANNER: A Tool to Develop Classical Planning Domains and Visualize Heuristic State-Space Search. We decided to share our code through Github: at WebPlanner, while the working app is available as Heroku app at WebPlanner.
ν-BDI (nu-BDI) is an interpreter for a normative agent that uses constraints on plan instantiations to enforce normative stipulations of obligations and prohibitions.
This interpreter is the result of the work I have been carrying out with Nir Oren, Wamberto Vasconcelos and Michael Luck, most of which is currently unpublished.
The basic work on this interpreter can be found in the paper Using Constraints for Norm-aware BDI Agents
It is implemented as an extension of the popular Jason interpreter.
I decided to share my code through Github:
Nu-BDI
EmPlan is a planner I implemented during my MSc for my planning BDI agent, while JavaGP is a recent Java reimplementation I made of the same algorithm for integration with Jason. After having noticed that there are no full Java implementations of Graphplan around on the internet, I decided to share my code at JavaGP@Github.
HTN2MDP is a probabilistic planner that takes as input a deterministic HTN domain, described in a subset of the popular JSHOP HTN domain description language . It generates a policy that tells, for each possible state in the domain, what is the best action to take. This planning consists of converting the HTN domain into a stochastic planning domain in the MDP format, and planning using an MDP solver.
The source code and binaries are available for download below. The package includes the binaries for Jason 1.3.1 as well as AIMA Libraries (for the MDP solver).
HTN2MDP - src
ANTICO is a proof-of-concept demo of a proactive information assistant I helped implement while at Carnegie Mellon University. It uses plan/intention recognition and probabilistic plan prediction to detect a human user's future information needs in order to provide time-sensitive information and guidance. The demo itself is intended to help a disaster response coordinator to go through all necessary steps to react to natural disasters and/or terrorist attacks.
Peleus is an implementation of planning capabilities within the AgentSpeak architecture, also known as AgentSpeak(PL) as published in two DALT papers: Composing high-level plans for declarative agent programming, and Leveraging new plans in AgentSpeak(PL). AgentSpeak consists of an agent architecture following the tradition of PRS agents, which define agents with procedural reasoning capabilities. Procedural reasoning is a mechanism used by agents to improve their runtime efficiency by limiting the agent's choices to concrete plans of actions that are adopted whenever a certain well-defined situation is detected by the agent. This constrains the agent's flexibility at runtime, since a designer has to foresee every possible relevant situation and define complete adequate plans for the agent to react to them. Peleus allows an agent to reason over declarative goals (as well as procedural ones) and generate plans to satisfy them by composing lower-level AgentSpeak plans. To compose these plans, Peleus infers the preconditions and effects of AgentSpeak plans and convert them into a STRIPS representation which is handed down to an external planning module. In this version of Peleus, we allow a designer to choose between the same planning process used by X-BDI, or an open-source Java implementation of Graphplan available at SourceForge (JavaGP). Peleus was implemented as an extension to Jason, which is a Java implementation of AgentSpeak. The name of this implementation is to conform with their AgentSpeak implementation, namely Jason from the Greek tale of the argonauts.
The source code and binaries of Peleus are available for download below. The package includes the binaries for Jason 0.9.3 as well as JavaGP, and EmPlan.
Peleus - src
This is a utility package including an implementation of a Jason environment driven by a script to generate new perception events at specific times during the execution of a Jason multiagent system. Class ScriptedEnvironment takes as a parameter an XML file with nodes containing a list of perceptual updates for all agents and the time at which these updates must be sent to all agents in the environment. This script file follows the schema available at www.meneguzzi.eu/schema/jason-script.xsd, an example of usage of this script is available at www.meneguzzi.eu/schema/script.xml.
The source code and binaries are available for download below. The package includes the binaries for Jason 1.2
ScriptedEnvironment - src bin
Iovis is an implementation of an AgentSpeak interpreter that includes a meta- level reasoning action library, and plans to process and comply to norms, following the technique from the paper Norm-based behaviour modification in BDI agents. With Iovis, a developer can deploy agents that change their plan libraries in reaction to norms from the environment, or from other agents (through electronic contracts, for example).
The source code for Iovis is available for download below. The package includes the binaries for Jason 1.2 as well as Jason Scripted Environment.
Iovis - src
Kastor is an implementation of an AgentSpeak interpreter extended with plan patterns for the creation of distributed plans following the technique from the paper Interaction among agents that plan. In Kastor, a developer can define external plans for each agent through the EP plan pattern, which creates a communication endpoint in the sharing agent allowing other agents in the society to request this plan to be executed on their behalf.
The source code and binaries of Kastor are available for download below. The package includes the binaries for Jason 1.2 as well as Peleus, and Jason Scripted Environment.
Kastor - src bin
Orpheus is an implementation of an AgentSpeak-based client for the classic version of the Trading Agent Competition (TAC). At the moment the implementation is a rather simple adapter pattern bridging a Jason agent to the reference Java AgentWare library provided by SICS to do the low-level communication with the TAC server. This, as well as other technical limitations, should be addressed some time in the future. As with Peleus, this software was named to conform with Jason's argonauts.
The source code and binaries of the TAC agent are provided below, the package includes the binary package of Jason, as well as the binary package of AgentWare.
Orpheus (for Jason 0.8) - bin/src
X2-BDI and Agent Viewer are two programs implemented as part of the research conducted throughout my masters course in order to verify the viability of the theories defended in my masters dissertation. X2-BDI is an extension to the X-BDI agent model/interpreter created by Michael Móra in his Ph.D. thesis, it includes a C++ implementation of the Graphplan planner which is used by the agent in order to perform means-ends reasoning. Agent Viewer is a Java tool that provides a graphical interface intended to ease interaction with X2-BDI, which was originally done through a Prolog console.
The implementations of X2-BDI and Agent Viewer are available for download below, as well as the C++ implementation of Graphplan used in X2-BDI and compiled as a windows dynamic library. The sources are also available for compilation in other platforms.
X2-BDI - src/bin (Requires SICStusProlog version 3.8.3)It is worth mentioning that these are experimental software programs that were not subjected to any formal testing process. Usage of the programs or its sources is free of charge for non-commercial use, otherwise it is necessary to contact the author (i.e. me).
DocDB is an extremely simple program that detects duplicates of files within a specific local folder. It was created to help me solve the problem of multiple versions of papers I download in the course of "gathering" research papers throughout the year for my PhD. Though I always try to be disciplined and keep a folder structure just for papers, and I try to categorize the PDF/PS files as I download them, as the number of files I download increases it becomes harder to remember things I downloaded or not, and I end up saving the same paper under different names in different folders. As another helper for this chore of bookkeeping research papers I use JabRef, which is a brilliant program to manage a BibTeX reference database. Since DocDB cannot even be qualified as alpha-quality software, user discretion is advised. Even though it is unlikely that any critical bug is present given the simplicity of this software, I recommend backing up the target files before executing DocDB.
The source code and binaries for DocDB are provided below. No build files have been created, but Eclipse project files are included in the source code, making it simple to compile and recreate the executable JAR file using Eclipse's interface.
DocDB - src bin
Not strictly software, but rather a Mac OS X workflow that adds a context menu item in Finder that when invoked opens an iTerm terminal tab, cd-ing to the selected directory. It is rather neat and useful if you use a lot of terminal applications in the Mac, and if you, like me, does not like Terminal.app too much. This workflow was inspired by Jeroen Leenars's workflow that opens Terminal.app instead of iTerm. Most of the code was unashamedly poached from an old macosxhints forum, attesting to my relative ignorance of Mac OS X scripting.
The workflow is provided below, and is provided as is, I do not guarantee it will work, even though it has worked in my Snow Leopard installation. It includes installation instructions, which are basically the following: open the zip file, go into the open-here folder, and copy Open iTerm Here.workflow to ~/Library/Services.
Open iTerm Here