I was surprised to see that an article I coauthored was finally published in the latest issue of Robotics and Automation Magazine. It had been so long ago that it took me a while to place the paper -- I guess it takes quite a while for a special issue to get published.

Anyway, the paper can be found here.

I was surprised to see that an article I coauthored was finally published in the latest issue of Robotics and Automation Magazine. It had been so long ago that it took me a while to place the paper -- I guess it takes quite a while for a special issue to get published.

Anyway, the paper can be found here.

A while back I submitted a short research outline to the AAAI Fall Symposium. It went over well enough that I have been expanding what I wrote there to continue towards my dissertation. Unfortunately, it did not scale directly; or rather, I need to restructure quite a bit to get it to scale up. So I have fallen to try to mind-map how all the bits and pieces inter-relate. I think that it has worked out rather well.

The mind-map doesn't really translate directly to an outline though. So I fell back on Tinderbox agents to help with this task. First I added a couple of simple attributes: Section and Order. Then we can use an agent to collect all the notes for a section, sorting on the Order attribute:

Since the agent is looking for notes with the section attribute that includes a given tag, we can place a note in multiple sections. The upshot is that I can now shuffle sections around just by changing two attributes and the agents automatically restructure the outline.

The downside is that I can only do this for a single level -- children of agents, but not grandchildren.

I just heard from Lefteris that he just got back from presenting his paper (well, our paper technically, but he did all the writing) at the 2005 Mediterranean Conference on Control and Automation. I've updated the link to reflect the final version in the proceedings.

I think I need to start submitting to these fun overseas conferences. The AAAI Fall Symposium in Alexandria, VA will be fun and it will be good to see my Grandma, but there is something about going to Cyprus for a work conference...

I'm attempting to write an extended abstract for Roles '05. Every so often I have to remind myself of a very useful tool that help me keep focused on the main points of the paper. These are from George Heilmeyer when he was the director of ARPA and are good to answer for any research project:

• What is the problem, why is it hard?

• How is it solved today?

• What is the new technical idea; why can we succeed now?

• What is the impact if successful?

• How will the program be organized?

• How will intermediate results be generated.

• How will you measure progress?

• What will it cost?

Ok, that should be all the papers I have been a co-author on to date. Expect a few coming down the pipe: I am working on a journal version of the thesis. Robin and Kimon are working up some feedback -- I probably have one more review cycle before it is ready to submit. I will also be working up a dissertation proposal. I will post that once it is baked.

A preview: The thesis was focused on a distributed architecture for heterogeneous robots. It used Jini as the middleware layer, and was almost entirely written in Java. As such Jini service entries carried generic information about robot state. Or rather they were posted into the registrar to enable capability-based search for services. We were also able to hook KAoS from IHMC in to provide some software-agent services. But this was done mostly in an ad-hoc way. I want to extend this to use a more complete ontology for the domain and then define policies in KAoS that would affect task execution on the robots in the system through constraints, obligations and authorizations. So that is the general idea -- just need to figure out the angle of attack and how to place this relative to other work.

A. L. Nelson, L. Doitsidis, M. T. Long, K. P. Valavanis, and R. R. Murphy, “Incorporation of MATLAB into a Distributed Behavioral Robotics Architecture,” 2004 IEEE/RSJ International Conference On Intelligent Robots And Systems (IROS04), Sept. 28 - Oct. 2, 2004, Sendai, Japan. [pdf]

Abstract—This paper presents a method that integrates MATLAB into a distributed behavioral robotics architecture. The architecture is written in Java and uses the Jini platform for distributed object registration, lookup and remote method invocation. The method described here can be used to integrate MATLAB into any Java-based behavioral architecture. The form of the integration allows a running MATLAB workspace to be accessed as a distributed object within the larger Java/Jini-based architecture. This is beneficial because MATLAB scripts and functions may be called in interpreted form and can make full use of MATLAB tool boxes and have access to the MATLAB workspace environment. This is not possible when MATLAB scripts are compiled into stand-alone C++, Java or p-code. The use of the architecture is demonstrated on an iRobot ATRV-JR robot and remote computer workstation. Experiments have been conducted to quantify GPS and odometry errors in outdoor environments using automated methods supported by the distributed architecture.

L. Doitsidis, A. L. Nelson, K. P. Valavanis, M. T. Long, R. R. Murphy. Experimental Validation of a MATLAB Based Control Architecture for Multiple Robot Outdoor Navigation, Proceedings of the 2005 International Symposium on Intelligent Control 13th Mediterranean Conference on Control and Automation, 2005. In publication. [pdf]

Abstract— Design, implementation and experimental validation of a MATLAB based autonomous robot control framework is presented, supported by, and integrated into a distributed field robot architecture known as distributed-SFX. The MATLAB based framework is composed of multi sensor fuzzy logic robot controllers that utilize laser, GPS and odometer data, fusing such sensor data and filtering out noise, to improve collision free navigation. Extensive outdoor environment experiments with single and multiple mobile robots are performed to demonstrate waypoint and goal point navigation, and raster scan search patterns in unknown environments with static and dynamic obstacles. Results and videos are provided to justify the proposed approach.

Long, Matthew T., Creating a distributed field robot architecture for multiple robots. Master’s thesis, University of South Florida, November 2004. [pdf]

This thesis describes the design and implementation of a distributed robot architecture, Distributed Field Robot Architecture. The approach taken in this thesis is threefold. First, the distributed architecture builds on existing hybrid deliberative/reactive architectures used for individual robots rather than creating a distributed architecture that requires re-engineering of existing robots. Second, the distributed layer of the architecture incorporates concepts from artificial intelligence and software agents. Third, the architecture is designed around Suns Jini middleware layer, rather than creating a middleware layer from scratch or attempting to adapt a software agent architecture.

This thesis makes three primary contributions, both theoretical and practical, to intelligent robotics. First, the thesis defines key characteristics of a distributed robot architecture. Second, this thesis describes, implements, and validates a distributed robot architecture. Third, the implementation with a team of mobile ground robots interacting with an external software “mission controller” agent in a complex, outdoor task is itself a contribution.

The architecture is validated with three existence proofs. First, an example is presented to show the implementation of a basic sensor service. Second, a basic behavior is presented to validate the reactive portion of the architecture. Finally, an intelligent agent is presented to validate the deliberative layer of the architecture and describe the integration with the distributed layer.

Read a little bit about ensembles and ensemble creation for class. [pdf]. The point behind ensembles is to create a number of classifiers, then evaluate an instance on all classifiers. The overall output of the ensemble is some function of the individual results, and is (hopefully) of a higher accuracy. The trick is creating the ensemble.

The paper describes a comparison of bagging (bootstrap aggregation) versus boosting and several random subspace / random tree / random forest variants.

The big point of the paper is that no one method is a sure-fire win versus bagging. Random-forests are significantly better sometimes, but have a significant loss (in the statistical sense). However, one key feature of the random-forest algorithms is that they are very fast, since they pick a random subset of features for evaluation. This can be a big win on data sets with large feature vectors. Genomic data, for example, has 65,000+ features -- it is computationally very slow to evaluate all features.

All that said, it is worth looking into building ensembles of classifiers to improve performance.

"Distributed Multi-Agent Diagnosis and Recovery from Sensor Failures," Long, M., Murphy, R., and Parker, L., also appears as IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS), Vol. 3, pp. 2506-2513, October 2003. [pdf]

Abstract—This paper presents work extending previous research in sensor fault tolerance, classification, and recovery from a single robot to a heterogeneous team of distributed robots. This approach allows teams of robots to share knowledge about the working environment, sensor and task state, to diagnose failures and also communicate to redistribute tasks in the event that a robot becomes inoperable. Our work presents several novel extensions to prior art: distributed fault handling and task management in a dynamic, distributed Java framework. This research was implemented and demonstrated on robots in a lab environment performing a simplified search operation.

"Affective Task Allocation for Distributed Multi-Robot Teams", A. Gage, R. Murphy, K. P. Valavanis, M. Long also submitted to IEEE Transactions on Robotics. [pdf]

Abstract— This article presents a novel emotion-based recruitment approach to the multi-robot task allocation problem. This approach requires less communication bandwidth than auction methods, enabling it to scale to large team sizes, and making it appropriate for low-power or stealth applications. Affective recruitment is tolerant of unreliable communications channels, and can find better solutions than simple greedy schedulers (based on experimental metrics of the time necessary to complete recruitment and the total number of messages transmitted). Experimental results in simulation and on three UGVs and one UAV in a mine-detection task show that affective recruitment succeeds with network failure rates up to 25% and requires 32% fewer transmissions compared to existing methods on average. Affective recruitment also scales better with team size, requiring up to 61% fewer transmissions than a greedy instantaneous scheduler that has an O(n) communications complexity, without a significant increase in allocation time.

"Validation of a Distributed Field Robot Architecture Integrated with a MATLAB Based Control Theoretic Environment: A Case Study of Fuzzy Logic Based Robot Navigation", K. P. Valavanis, A. L. Nelson, L. Doitsidis, M. Long, R. R. Murphy also submitted to IEEE Robotics & Automation magazine. [pdf]

The paper presents fundamental aspects of a multi layer, hybrid, deliberative and reactive Distributed Field Robot Architecture (DFRA) that has been designed to support functionality of heterogeneous teams of unmanned (ground and aerial) robot vehicles. The DFRA is implemented in Java using Jini to manage distributed objects, services and modules between robots and other system components. It is interfaced with a control theoretic MATLAB environment, which is supported and integrated into the Java based framework using the JMatLink Java class library. This allows modules and services implemented as native interpreted MATLAB code to be accessed as remote and distributed objects. The combination of the Java based distributed architecture and the use of MATLAB in its interpreted form for autonomous robot navigation and control is a unique aspect of the reported research.

Experimental validation of the DFRA and its MATLAB integration is demonstrated by implementing simple prototype support modules for robot navigation. These modules include: i) a time-history laser filter module; ii) a heuristic GPS-based pose detection module; iii) fuzzy logic controllers that utilize laser, GPS and odometer data as inputs. Navigation experiments in the field utilize single and multiple robots and included scenarios in which a single robot navigated through an environment with many unknown obstacles to reach a distant goal location, and scenarios in which robots executed search routines by traveling through sets of way points. Robots negotiated both static obstacles and dynamic obstacles including other robots.

"Distributed Error Handling and HRI," B. Zimmel, M. Long, J. Carlson, R. Murphy, also to appear in 2004 IEEE International Conference on Robotics and Automation (ICRA), 2004. [pdf]

Abstract—The implementations of a distributed, autonomous error handler (EH) and a human-robot interface (HRI) are presented. The interface is combined with the EH to allow a human operator to see that a failure has occurred on a robot and whether or not it has been served by the EH. An experiment was run to test how well the EH and the interface work together, as well as the usefulness of the EH. The results were inconclusive, although the EH and interface worked together successfully.

"Application of the Distributed Field Robot Architecture to a Simulated Demining Task", Matt Long, Aaron Gage, Robin Murphy and Kimon Valavanis also to appear in "2005 IEEE International Conference on Robotics and Automation (ICRA), Barcelona, Spain". [pdf]

Abstract—As mobile robot teams become more complex, it is necessary to develop a control architecture to manage the resources present in the team. The Distributed Field Robot Architecture (DFRA) is a distributed, object-oriented implementation of the SFX hybrid robot architecture that allows for dynamic discovery and acquisition of robot resources and the seamless integration of humans and artificial agents in the robot team. This paper introduces the DFRA and details its application to a high-fidelity demining scenario using a heterogeneous team of ground and aerial robots.