Apr 10, 2010
Class
Alex Brown - "Leaf Microcontroller board"
Alex gave a lecture on the microcontroller board he designed for the Leaf project. You can find out more about this board on the Leaf website. This board has room for an HC9S12DP256 microcontroller daughter card and various sensors, including gyro, accelerometer and compass. When designing this board, he was looking for a good connector to hook up the external sensors and actuators. The connector had to be easy to assemble (i.e. connect wire to connector), easy to insert/remove, remain fixed under moderate vibration. He settled for a female PCB-mounted phone jack.
Alex mentioned that all sensors require 5V. But when supplying a single 5V source to all sensors creates noise. Instead he suggests supplying 8V. Each sensor regulates it down to 5V. This reduces the noise. The servo's require 6V.
Business
upcoming classes
May: "robot sensors" by Thomas Messersmidt
June: "Leaf & List" by Bruce Weimer
July: "Lisp 2" by Bruce Weimer
upcoming contests
June: Hallway competition
August: RoboMagellan
Treasury: $427
Bruce visited the Pomona RC Fair. There he met Erin who is a master RC plane and helicopter pilot. He flies a T-REX 600 Nitro R/C helicopter.
He is available for a modest fee ($100) to give a demonstration at the RSSC meeting. Given the budget of the club, this price was deemed rather high. Also, the University has strict rules against RC planes on campus. Given that we get the meeting room for free, we do not want to do anything to jeopardize our good relation with the school.
As a consolation, here is a video of his superior pilot skills:
Jef Mangelschots announced the possibility of a lecture by Hanno Sanders from hannoware.com in New Zeeland on Propeller development tools. The lecture would be given on the 4th saturday meeting. That is when Hanno arrives at LAX. There is a chance that the guest lecturer will not make it in time.
Jef Mangelschots presented the results of his solicitation for donations from robotics-related vendors. These donations can be found here. Purpose is to use the donations as prizes in our upcoming competitions. This will give incentive to our members to participate.
Contest
This month we held a pick-up-the-can contest. Wanna know more, then follow this link ...
Show & tell
Don Fears showed us a low-cost robot he built out of foamboard, some toy wheels and plastic gear motors. All for $24.
Jim bought a toy electronic drum kit at AJ Wright for $15. He is interested in it's sensors for detecting taps. They seem to be brass discs and seem to be piezo-electric somehow. He was wondering how to use them as tactile sensors.
John Walters showed us his work in progress on his hallway competition robot. He is working on a scanning IR rangefinder sensor to detect walls.
Alex Brown' eternal quest for the ideal robot base brought him to experiment whith a rectangular base with two castor wheels (instead of the usual one) and 2 driven motors. He demonstrated how castors, when driven against their orientation cause a jerk, that throw the robot off course.
Alex was wondering about a configuration that eliminates that effect.
Jef Mangelschots suggested a dual-wheeled mobile robot configuration developed by McGill Centre for Intelligent Machines, where 2 motors pointed downward drive a wheel each, and provide a type of Ackerman steering.
Martin Mason talked about his work on getting robots to understand human commands. It is reasonably straightforward to develop a robot that understands spoken commands and train it on a small instruction set. Downside is that the user has to remember the set of commands and use them in the right order and only use the predefined combinations. In many cases, the speech recognition software misses a word or interprets it into another word, leaving the robot baffled as to what the user wanted it to do. The effect is a little like talking to someone over a cell phone with a bad connection. Many times, words disappear, and yet many humans are able to reconstruct what is said in their head.
To make voice-operated robots more user-friendly/fault-tolerant, he worked on algorithms that give robot the same capability to reconstruct the correct intent from a multi-word spoken command of which individual words are missing or misinterpreted.
In his experiment, he has a 2m x 2m playing field with a omniwheeled Robotis-based robot bunny. Various colored baskets are spread over the field. An overhead camera footage is streamed through OpenCV module to detect the position of the 'bunny' and the baskets. The purpose is for the user to say commands like "please move the red basket to the yellow basket" or variations on that. More about that work here.
He compared both Roborealm and OpenCV, and found Roborealm to be about 40% faster on a Windows PC.
Unfortunately Roborealm can not be used on Embeded Linux platforms, liek the Beagleboard. OpenCV is better suited for this. Here he describes how to build OpenCV for Beagle board. Here is an article in which Martin shares some thoughts on compatible webcams.
He used Roborealm to identify corners, baskets by color and their coordinates. The robot was identified by a fiducial. More on that here.
He learned the robot how to pick up the basket by pre-recorded poses.
This is a link where he discusses how to move a humanoid around, based on tracking an attached fiducial from an overhead camera.
The easiest and cheapest way of voice recognition is the Microsoft SAPI. Here you can find more about using SAPI from Python.
An alternative speech API is NUANCE Dragon SDK.
Martin then went on and talked about robot path planning. He presented a Binary divide & Conquer algorithm that subdivides the image overlaying a terrain in 4 equal pieces. He continues to subdived the pieces that have the least obstacle-pixels until there is a path of empty squares between robot and end coordinates. Alternative approaches are:
- Wavefront
- Dijkstra path finder
- A*
- STRABO Pathfinder
Henry Arnold answered a call for volunteers to help out start a robotics class for the Orange County chapter of 4-H.
He started out with an Escape Robot kit (approx $25).
He then developed a PICAXE-based microcontroler board (cheaper than a BASIC STAMP). He implemented an interface to drive the robot remotely with the Lynxmotion PS2 controller (approx $20).
Kelly Mattis showed us his WIP on his robot R.A.L.P.h. Is is made mostly out of PVC components from the hardware store. Unfortunately, he just fried his motherboard (intends to use Ubuntu Linux) before the meeting ("It worked yesterday" strikes again !). Good luck with your endeavour. Please keep us posted.
It was also mentioned that Roborealm has a faceless mode, Kiosk mode and can run over a TCP/IP socket.
It also has Python scripting capability.