Portland Area Robotics Society

Regular Meeting Minutes

June 3, 2006, 10:30AM



1.      Welcome and Announcements

2.      BotFest Report

3.      Presentations:

a.       Prof. Scott Wallace: AI Agents

4.      News Reports

a.       Show-and-tell



BotFest Report

BotFest was very successful. Attendance was excellent, including a school class from Brookings. Many competitors expressed interest in future events.


We had many competitors in Mini Sumo, most of which were students from Brookings. We had few competitors on some of the newer and more technical events like Advanced Sumo (only one robot), Japan class Sumo, Maze Solving, Walker, and RoboMagellan.


Dave Shinsel, as the only competitor in RoboMagellan, spent a great deal of time with the crowd to explain his robot.


There were 3 Talent Show entries: 2 holonomic robots, and one bipedal Robonova.


Competitor fees and concessions yielded enough money to nearly cover the cost of the venue.


Thanks to all volunteers and organizers. Thanks to BC LaserWorks for donating the award trophies (Laser cut in clear lexan) %G–%@ www.bclaserworks.com.



Prof. Scott Wallace: AI Agents

Scott Wallace teaches at WSU Vancouver. Grad training in AI lab at Univ. of Michigan. Current research is in Agent technologies. Teaches an AI class at WSU.

Slides will be made available on-line.


AI software can help with common robotics concerns:

1. Route Planning

2. Reactive Behavior via Rule based systems

If the environment changes, the agent has to react and change its course of action.

3. Deeper deliberation through planning

Determining appropriate, efficient order for executing tasks.

4. Probablistic models of actions.

Dealing with uncertainty. I.e. the interactions of hardware systems and software; the model the robot uses in decision making.


Getting from A to B efficiently -- Graph-based route finding.

A. Obvious approaches

1)      Check in expanding concentric circles %G–%@ Bread-first search. Tends to find the route in terms of the fewest number of hops, but not necessarily the cheapest route.

2)      Depth first search %G–%@ keep following a single path until it ends. Back out of mistakes until the target is found. May find the goal through a very inefficient route. Have to track visited nodes to avoid loops.

3)      Uniform Cost search: Track cost incurred and track the cheapest solution. Use an approach similar to depth-first search. A route is preferred if its cost is lower than alternatives.

B. Heuristic approaches

1)      A-star search: Use knowledge of position to bias the search. Coordinates of destination and current position to estimate distance remaining to the target. The route search incorporates the cost of the current route and the cost of the estimate of distance remaining. This biases the search toward the most efficient route.

Reacting to a changing world

A.   Sense, Think, Act

1)      We want robot agents to react appropriately when the situation changes in unpredicted ways.

B.   Hierarchical Decomposition

1)      Set goals based on current situation, but retain the ability to sense, think and react based on changing environment.

2)      Programming is broken into small pieces that are interruptible based on new circumstances.

C. Rule-based systems

1)      Rule-based systems encode knowledge in if-then rules.

2)      Rules can process sensory information, propose actions or goals, evaluate

3)      As conditions change, irrelevant goals automatically go away

4)      Example rule-based system: Univ of Michigan SOAR

Where am I? %G–%@ Dealing with uncertainty.

A.   Robot actions are not completely deterministic

Simple behavior of moving forwared may actually result in some sideways movement. This makes dead-reckoning of position error-prone.

Probablistic models of action

If we know what results are possible given a desired action, and we know their probabilities, then we can create a probabilistic model of our state. i.e. an algorithm that computes the robots most likely position.

Coupling a probabilistic model with an external source of information can increase certainty of position.

B.   Sensor to symbol problem.

Turning sensor data into data that can be used by AI systems.


Pyro (python robots): http://emergent.brynmawr.edu/~dblank/pyro/

Soar (rule based system): Http://sitemaker.umich.edu/soar

Sensory Graphplan (planning): http://www.cs.washington.edu/ai/sgp.html


Book: AI A modern approach by Russell and Norvig


Prof. Scott Wallace web site: http://ai.vancouver.wsu.edu/~wallaces

Show-and-Tell and News

Jeff S. displayed Spartan3 FPGA board and discussed simulation of AVR microcontrollers.


Steve D. demonstrated two VEX robots. One holonomic robot platform, and another 6-wheel rover. He is mixing Erector set parts into his designs. Several versions of Visual C exist for programming vex bots. Steve recommends Make magazine. The latest issue discusses the MAKE microcontroller, which will sell for $150 with support for real-time operating system programming and easy motor control.


Monty G. displayed his work-in-progress maze solver based on a BDMicro AVR controller board.


Paul B. brought in some new mini and micro sumos. He recommends a Piranha battery charger, available at RC supply shops.


Mark M. had a couple of boards fabd at PCB Express for his Maxwell robot.


The TI EASY 430 was recommended as a simple cheap MSP development system.


Pete S. is working with ATMEL ARM controllers for his work projects. Pete recommends reading www.robots.net for interesting robot news. Pete suggests a new competition for our club: Funny Golf is a competition between two robots with colored holes and balls.