Robotics References

Reference:

Brooks, Rodney. "Achieving Artificial Intelligence through Building Robots." Boston: Massachusetts Institute of Technology, 1986.

Application:

Rodney Brooks warns that abstraction can be dangerous and that truly intelligent robotic systems will not follow the box world we have created for them. Therefore, we should not design our robot with limited conceptions of its world, but instead create the desired complex behavior by combining simple behaviors. Unfortunately, our robots are not of the right architecture to create "robotic insects," but in some cases using a subsumption architecture may be appropriate.

Reference:

Horswill, Ian. "The Polly System." AI and Mobile Robots.

Application:

The Polly system has some similarities to our own robot in that both are intended to use vision to perform tasks for extended periods of time. Both robots also try to find humans. The Polly system uses place recognition that is different to ours. Polly is sensitive to dynamic changes in the environment, while our system relies on navigating by fixed landmarks such as doors and hallways.

Reference:

Vaughan, R. N. Sumpter, A. Frost, and S. Cameron. "Experiments in Automatic Flock Control." Edinburgh, UK, 1998.

Application:

This look at robot-animal interaction, while interesting, was also very simple. Our experiment will be involving humans, which, unfortunately, don't exhibit flock behavior. At least, they don't usually. Additionally, the experiment "cheated" in that an overhead camera was used to help coordinate the positions of the actors (the robot and the ducks) instead of using only sensors on the robot itself. We will not have this option for our experiment and so very little of this experiment usefully applies to our experiment.

Reference:

Mark Yim, David G. Duff, and Kimon D. Roufas. "PolyBot: a Modular Reconfigurable Robot." IEEE International Conference on Robotics & Automation. April 2000.

Application:

While fascinating in and of itself, PolyBot holds very little relevance to our project. The paper is on building modular reconfigurable robots, a type of robot architecture that our robot is not.

Reference:

Martin, Martin C. and Hans Moravec. "Robot Evidence Grids." CMU RI TR 96-06, 1996.

Application:

Robot evidence grids provide an algorithm by which to map the robot's environment. Given sufficient time to explore, they can provide accurate mappings of obstacles, even from low quality sensors. Since it is likely we will also be using low quality sensors, we may consider implementing this algorithm into our robot's "brain."

Reference:

Moravec, Hans. "Robots, After All." Communications of the ACM. October 2003. Vol. 46, No. 10.

Application:

This paper contains very little implementation level details of robotics, but instead steps back to give a broad view of what the future may hold. The paper holds very little application to our project specifically, but may be applicable to us over time and as we look for changes in robotic technology over the next several years.

Reference:

Application:

Use of modules, percept pair, simulators are not sufficient, "a highly detailed model of reality is doomed because of enormous computational complexity", state-set representation, interleave planning and execution, must be undo-able,

Reference:

Application:

We are using Monte Carlo Localization (MCL) on our Evolution to efficiently and effectively map it's position within the Libra Complex (we already have a map).

Reference:

Thrun, Sebastian. "Robotic Mapping: A Survey." CMU-CS-02-111, February 2002.

Application:

(Insert applications comments here)

Reference:

R. Grabowski, L. Navarro-Serment, and P. Khosla. "An Army of Small Robots." SciAm Online May 2004.

Application:

While interesting, this article had very little relevance to our project. We are working with the relatively large Evolution platform, not a several centimeter "mini". Also, we only have one robot, so group mapping/localization techniques and such do not apply. Neat ideas, but not directly applicable.

Also, with a title including both the words "Army" and "Robots", the lack of an in-depth discussion of equipping rocket launchers is simply unacceptable.

Reference:

Application:

Naturally, our Big Brother Bot program will incorporte some degree of Anti-failure robustness. Post-failure robustness, however, will also be crucial. There are two primary cases where post-failure recovery is not only possible, but expected by any reasonable tracking robot. The first is if the robot gets lost (e.g. the kidnapping problem). To handle this, we are using MCL which naturally incorporates a degree of post-failure robustness because it can address both the kidnapping and the localization problem at the same time and by using very similar methods. The other major failure will be losing track of the target BBB was tracking. For this situation, we will probably use a simpler version of Toyoma and Hager's Incremental Focus of Attention (IFA) framework. We will first try to re-find the target using the camera, but if that fails, the robot will resort to its wandering algorithm looking for another target. This is a form of post-failure recovery.

Reference:

Application:

While the RRT-Connect is a fascinating and attractive path-planning algorithm, our sentinel robot did not use path planning. If an obstacle comes between it and its prey, it merely stops. A possible enhancement to our robot would be to add a complex path-planning algorithm, such as RRT-Connect, thereby enabling the robot to pursue and destroy its target more easily.