Jan 09, 2010
Class
Singing Leaf robot - Thomas Messersmidt
Thomas demonstrated the singing capabilities of his Betty9 robot using Vocaloid software (about $90)
He also talked about servo motors. In his experience, ALL manufacturers LIE about the torque specs. His suggestion is to predict the torque requirements and select a motor of which the advertised torque is not lower than THREE times the required torque.
Business meeting
Upcoming contests
February: sumobot
March: Line Following
April: Pick-up-the-can
May: Hallway contest
Upcoming classes:
February: "Introduction to AVR Programming" - Jef Mangelschots
March: "Image processing with Blaze" - Robin Hewitt
April: "Leaf microcontroller" - Alex Brown
May: "Introduction to LISP Programming in Leaf" - Dr. Bruce Weimer
September: "Machine Intelligence & Neural networks" - Martin Mason
Business
Robin Proposed she would develop a website where people can give credit to other people for influencing their robotics projects.
Mt Sac College is holding a robot fair in October
It was proposed to move to cheaper web-hosting services. One-Dollar hosting was suggested.
Show & Tell
Alex Brown talked about his experience with GPS for navigation of indoor Leaf-type robots. Follow this link for a detaled coverage of that discussion.
Thomas showed an idea for a directional microphone for voice-operated robots. He uses a plastic bottle with a removed bottom, to wrap around a USB microphone:
Martin Mason, who visited us from his yearlong sabatical at Plymouth University, talked about several topics that were worthy of his attention in the preceeding months:
He studied bipedal walking robots using the Bioloid platform using AX-12 servo's (here is write-up on AX-12 servo response). He talked about static walking vs dynamic walking gaits in bipedal robots. Static walking gaits ensure the robot is always stable, even at any given time during a move. A dynamic walking gait, which is far more complex to achieve, is more of a controlled sequence of not falling, i.e. if a robot would stop in the middle of a move, it would fall over. Dynamic walking however, when implemented correctly, yields a much more natural walking motion. In order to achieve this, you need feedback mechanisms to correct any errors. Humans correct their walking using inner ear (equivalent of an IMU), visual clues and pressure indicators from the feet.
in order to finetune the recorded movements of his Bioloid robot, he experimented with various techniques for capturing stresses from the robot's foot soles to use as feedback for controlling the walking motion.
His first attempt was using strain gauge foot sensors. He attaches strain gauges, placed in a Wheatstone Bridge. The results were somewhat disappointing. The readings varied over time. THe link to his blog has a link to a detailed report.
Their second attempt is a foot sensor based on piezo-electric FSR (Force Sensitive Resistors). They do not use it continuously for walking, only for calibrating the walking algorithms.
He later added this update on foot sensors.
Martin also talked about adding IMU to their Bioloid robot. (here is his report on IMU data capture, here is more info on integration of the Bioloid IMU Gyro, here is a link to some IMU software he worked on).
He also showed us his work on robot arm control simulation and inverse kinematics to be able to move the arms such that a robot can scratch its own back without trying to move the arm through the body, but instead calculate a path around its own body.
The goal was to allow Bioloid Pose control.
He also offered this insight into robot walking.
Martin also briefly mentioned new Physics engines for use in robot simulation software:
- Newton
- Nvidia's PhysX engine
- He himself has had good results with ODE (Open Dynamics Engine) and its Python wrapper PyODE. Here is a blog on that. it's main problem is that is somewhat slow, compared to the newer physics engines.
Notable robot components:
VRbot voice recognition - it understands spoke commands in different languages, offers an API and it is able to upload new words.
Venus GPS module from Sparkfun
Martin and John Davis demonstrated OpenCV on an open source Beagleboard embeded Linux board. This is an ARM8 Linux board with HDMI video output. It uses an incredible low power (100mA @ 5V). It also has a separate DSP processor (TMS320C64x+) which allows 430MHz signal processing.
Here is more about OpenCV.
Here is more about adapting OpenCV to Beagleboard Linux.
John Davis was successful in improving the OpenCV frame rate, by incorporating SDL in OpenCV.
- Harris Corner finder is a technique of finding corners in frames to determine a flow vector that can be used in Optical Flow used in Obstacle Avoidance. When an object is close, it's flow vector is large, otherwise it is small.
- Here is another article.
- KLT is a C library that can be used for corner detection.
John Walter showed us his Hallway Competition robot: