I followed the build-from-source instructions for Desktop-Full, installing dependencies as they came up (i.e., when CMake complained) and invoking catkin_make_isolated with the following: -DCMAKE_INSTALL_PREFIX=/opt/ros/groovy -DPYTHON_EXECUTABLE=/usr/bin/python2 -DPYTHON_INCLUDE_DIR=/usr/include/python2.7 -DPYTHON_LIBRARY=/usr/lib/libpython2.7.so -DSETUPTOOLS_DEB_LAYOUT=OFF

Some of the most annoying oddities:

• camera_calibraton_parsers wants yaml-cpp 0.3 (from the AUR), not 0.5!
• image_proc and stereo_image_proc use Boost’s lock_guard in their nodelet code but don’t compile because the BOOST_VERSION check doesn’t work. I tried passing -DBOOST_VERSION=105300 to make to no avail. I just edited the source code. (Though in hindsight, maybe I should have passed that to gcc by setting CFLAGS in the appropriate CMakeLists.txt.)
• gscam needs gstreamer0.10-plugins installed.

The MOST annoying thing of all was waiting for catkin to sift through nearly 180 packages multiple times before choking up on various packages, over and over again. Even on eight hyperthreaded cores, this took a while. It would have been smarter to check ros.org for which packages each removed package was a dependency of and remove them from src preemptively. Even better, I could probably have used rosdep to generate and remove a list of dependent packages.

Initially, there was no pain and I vainly tried to pull it back myself, thinking it was a dislocation.

In retrospect, I should have kept my front wheel straight and leaned backward more. And maybe not tried to pull my finger back by myself.

What sucks more than not being able to bike for the next few weeks is that I can’t use my left hand for typing anymore. I have to supinate my arm uncomfortably to even hit a key with my index finger without mashing the keyboard with the cast. Doing any precision mechanical work is also going to be a challenge.

That said, I’m getting excited thinking of all sorts of mechanical additions to build onto the cast when I’m back in the lab.

Update 5/9/13: Got three pins inserted yesterday at 13:00. I was under general anesthesia and apparently woke up and talked with the doctors just after surgery was over, but I don’t remember anything. All of this is a first for me, so it’s interesting. I remember waking up at 14:50 and was thinking and talking coherently by 15:10.

Then they gave me hydrocodone and removed the IV (they have flexible needles!). The painkiller took full effect by 16:00, and I experienced the worst feeling of nausea I’d had in a long while. Got home and vomited three times.

The nausea didn’t go away until 6:00 today when I woke up. That didn’t leave much time for me to study for the midterm exam I just took, which actually wasn’t that bad. I’ll probably get at least a C.

I forgot to mention last time that the nurse used an oscillating saw to cut my cast open. An oscillating saw cuts through hard material like the cast but not the cotton or skin underneath, which is cool.

Innovation is all about change and thus takes discipline and willpower.

So stop being a lazy asshole and get stuff done.

DIY half-wave dipole antennas on 900 MHz XBee Pros.

I followed this tutorial to make the antennas.

Before I found the tutorial above, I tried cutting a piece of 20 AWG magnet wire to 8.33 cm (a quarter-wavelength), but that didn’t work, probably because the magnet wire didn’t have the right impedance.

I have since flown the tricopter as high as 200 ft directly overhead in the past seven months of usage with no connectivity issues.

There are differences between the cheap 1 mW XBee and the XBee Pro 900 in the configuration commands they take. The datasheet proved very useful.

RCGroups has another good tutorial on this topic, and this is an insightful discussion on RF.

It would be a good exercise for me to implement such neural networks in Lisp, both of which I am currently learning.

To make things even more exciting, I can train a program to (aurally) recognize appropriate pauses in people’s conversations (e.g., “umm”, “uhh”, “so…”) and politely chime in with verbal suggestions.

Since doing so audibly might become too intrusive, I might make some of the incoming ECEs build an LED matrix (or just buy one) on which the program can display the words.

This might become a software module for a more sophisticated AI bot I build in a future project.

UPDATE 10/22/12: The bot could also compose poems. And music. And draw things. In short, this bot could be generalized to complete anything.

Yet the ideas keep coming, and my todo list (a text file in my home directory) is hundreds of lines long.

Therefore, without further ado:

Idea 0: Post ideas!

I’ll try to post each of my ideas individually, along with related subideas, anecdotes, or elaborations, as appropriate. They may be good or bad, and other people probably have had the same ideas. Regardless, I hope to at least keep my todo list within the scope of a few weeks and months, not years.

Bootloader written!

In the scramble leading up to the IARC, I donated my Arduino’s atmega328p to the Aerial Team, which subsequently fried the chip.

I ordered a new chip from Digi-Key a few days ago, but without the Arduino bootloader, I couldn’t upload sketches to the board over USB.

I tried at first to use a Pololu PGM03A programmer I had lying around the lab, using avrdude from the command line. This failed to work, because avrdude kept seeing an invalid device signature.

My other option was to use the AVRISP mkII, which I had never used before because my Arch Linux installation had issues recognizing the programmer. This was a good chance to rectify the issue. The solution was simple:

I saved the following as /etc/udev/rules.d/60-avrisp.rules:

SUBSYSTEM!="usb_device", ACTION!="add", GOTO="avrisp_end"                            

# Atmel Corp. JTAG ICE mkII
ATTR{idVendor}=="03eb", ATTR{idProduct}=="2103", MODE="660", GROUP="dialout"
# Atmel Corp. AVRISP mkII
ATTR{idVendor}=="03eb", ATTR{idProduct}=="2104", MODE="660", GROUP="dialout"
# Atmel Corp. Dragon
ATTR{idVendor}=="03eb", ATTR{idProduct}=="2107", MODE="660", GROUP="dialout"

LABEL="avrisp_end"

and added my user to the group “dialout” (I had to create this group on my machine).

I could then use the AVRISPmkII to burn the bootloader through the Arduino IDE.

Tricopter folded for transport.

The aluminum arms are now foldable, and the motors are closer to the center of the chassis (I have yet to trim the arms to a shorter length). There is room for the new and bigger ESCs underneath the other electronics.

Electronics mounted with vibration-isolating rubber grommets.

The Seeeduino/protoboard/sensor blob is bolted to a wooden plate, which, in turn, is attached to the center piece with bolts isolated by rubber grommets. This, combined with a rigid mounting of the battery to the center piece, seems to work much better than foam in isolating the sensors from vibration.

It just happens to be able to balance on the tip of the tail arm.

Total weight of the machine, with battery, is 2050 grams.

The current (6th) mission of the IARC requires teams to develop an aerial robot capable of autonomously navigating an indoor space (i.e., hallways, furnished rooms) to locate and retrieve a flash drive, all the while avoiding or disabling laser tripwires, avoiding a security camera by its blue status LED, and optionally recognizing signs that help guide the quadrotor toward the room containing the flash drive.

Window mockup for computer vision testing with quadrotor.

We spent the majority of the past school year developing our own flight platform (electronics, hardware, and software), leaving only the last two months for real hardware and autonomous navigation development. Nevertheless, we have made significant progress in that time toward making our quadrotor capable of completing the mission.

Realistically, we may not be able to complete all requirements of the mission this year, but at this point our quadrotor can autonomously take off, maintain altitude, follow a wall, and land. I have started developing the computer vision system (using OpenCV) that will allow the quadrotor to locate the window for ingress and egress of the arena, recognize the blue LED of the security camera, and recognize the flash drive.

Multirotors and supplies half-packed.

After some last-minute tests, we packed up and left Corvallis yesterday at 13:00 PST. It’s 10:50 PST now, and we are 40 miles away from Grand Forks, ND, where the competition will be held.

Of the three Red Brick 70A ESCs I ordered from HobbyKing to replace my burnt-up ones, one had different pin assignments to the FETs than the other two. This required me to upload the rb50a firmware instead of the rb70a.

Different board.

Before I figured out that the rb50a firmware worked on this board, I spent a while fiddling with the rb70a firmware and found that I need to consider the ADC pin assignments as well as the FET pin assignments for the commutation to work properly. Otherwise, the ESC will beep at startup but fail to spin the motor because it will check the wrong pair of output leads during a given phase.

The calibration feature of simonk’s firmware threw me off for a while because I accidentally calibrated the ESCs to a ridiculously small PWM input range and didn’t realize it.