AWE

Edmund Fung Calif. USA

[adventureswithedmund]

Philips Hue light dimmer switch

“Ok Google, turn on the lights. Ok Google Ok G- Turn on the lig- *I can’t help with that right now* Damn it .. “

I bought the hue lights a while back so I can setup timers to turn on the lights and naturally wake me up in the mornings. The colors were cool too especially the deep red or sunset orange colors. Controlling them with my Google Home with voice was not as easy as I thought however.

Next novel idea? A light switch and dimmer! I had made this a while ago, but was still on a proof-of-concept breadboard. Today, I neatly packaged the switch and stuck it where it can be easily used.

Materials

Hue lights

ESP8266 - basically an arduino with wifi

Breadboard, wires, resistor (for debounce)

3D printed nob

Code: https://github.com/edfungus/hue-esp8266

Features

On - push quickly

Off - push and hold for more than 300ms

Dim - rotate count-clockwise

Brighten - rotate clockwise

It is fairly simple. I wanted to do everything with one button.

Connecting to Hue

Connecting to Hue is pretty simple since they have good API documentation and easy way to get an API key for your Hue (getting started guide). Unfortunately, the API is only via REST so there is a slight delay when cranking the brighten up or down.

(Pixel 2′s auto-exposure ensures that we can’t see the brightness change 😞)

Encoder/button

The main cool part of this project was writing the code to handle the encoder. The hardware I used also had a button on the encoder if you pushed down. An encoder was nice since it had unlimited turns and therefore wouldn’t “bottom out”.

I used a single interrupt to handle both of the encoder pins which incremented a value to be sent for brightness update. The encoder also needed to know its previous state in order to know whether the current state indicated a shift in position.

The button just used a saved state and the current time to know whether or not the button has been “long” pushed so it could turn off the lights.

Since we are using the ESP8266, sending a PUT request to the Hue bridge was dead simple.

Conclusion

Not as pretty, but fully functional. Or if you are super rich, you can just spend $25 for the Philips one below. But what fun is that?

[adventureswithedmund]

Carbon Fiber Cycling Shoes

Long time no see! Introducing some homemade cycling shoes!

I’ve always wanted to use carbon fiber to make something. It’s a pretty cool material and was less boring sounding than wood working and less costly than metal working. Find practical applications was actually more difficult than I thought. With my current cycling shoes wearing thin, I figure carbon fiber could reinforce a pair of soccer cleats making it a cycling shoe!

Carbon fiber can be used and processed in many ways. Because I had limited access to equipment and don’t have a bunch of fun money, I took some shortcuts and trade offs to make the carbon fiber cycling shoes work.

The Plan

The plan is to use a donor shoe as the upper, add carbon fiber to make the sole stiffer and add bolts for cleat attachment. I have only found one forum post where someone else attempted this so prior information was sparse. Found some cheap orange cleats on ebay so next was getting the carbon fiber materials. 

Using Carbon Fiber

Carbon fiber itself is a flexible cloth which is great to warp around things. Like cloth there are different weaves. I picked a 2x2 twill carbon fiber weave because it was more flexible for tight bends. To cure the carbon fiber, I used Max 1618 epoxy which did not require high heat curing temperatures.

Because I didn’t have an oven in which I can cure the carbon fiber nor did I want to invest in a vacuum pump, I opted to just let my carbon fiber cure in the garage while being tightly wrapped in saran warp. The trade offs from not using an oven is that the epoxy may not cure as hard or impregnate the epoxy as well. The trade offs for not using a vacuum pump is that the part may end up being heavier as excess epoxy will remain in the part.

I did research pre-preg carbon fiber which was essentially carbon fiber with epoxy already in it. While this was nice, pre-preg was quite expensive and also had a shelf life. Perhaps if I really knew what I was going to make with carbon fiber than I would opt for pre-preg carbon fiber down the line.

Regarding wrapping the shoe in saran wrap, I also used a layer of peel ply between the saran wrap and carbon. This allowed me to easily remove the saran wrap since it would not be in contact with the epoxy. The peel ply doesn’t stick to epoxy so that was easy to remove too. 

Preparing the Shoe

I used a Nike Mercurial as a donor shoe primarily because it looked cool. To create a smooth bottom, I removed the studs from the bottom of the shoe with a dremel and sanded down the surface. I kept the rear studs so they balance with a cleat on. 

To prove that this concept will even work, I measured out 3 holes for cleats with my current pair of cycling shoes. To receive the screws from the cleat bolts, I bought a Shimano cleat nut replacement part SH-R121 which is normally for a Shimano shoe. This part is amazing because it had the correct 3 screw spacing and curvature already set in one piece of thin metal! I took a test ride in the the proof of concept and they functioned! They felt a but sloppy because the bottom of the shoe was way to flexible but they did clip in and out on command :)

Carbon fiber could be added in a couple ways to the shoe.

I originally wanted to remove the sole of the shoe and cover that completely with carbon fiber. Unfortunately, I couldn’t get the sole off the shoe. I even tried several methods including boiling water but it wasn’t budging.

Another method was to create a carbon fiber insert inside the shoe. The great thing would be that the insole may be transferable and I wouldn’t damage the outside of the shoe. I needed to get a mold of the inside of the shoe with a material that also would durable and not be brittle. I came across epoxy clay which is amazing! I was able to model the inside of the shoe. The biggest issue I realized afterwards was that an insert, even replacing the original insole, would have taken up too much space inside the shoe. As a result, I never made the carbon fiber insert.

The most obvious method is to epoxy carbon fiber on to the bottom of the shoe. I prepared a template for the carbon fiber cutout but taping masking tape on the shoe and removing it. This way, the masking tape perseveres the 3D aspect of the shape. I cut out three pieces of carbon fiber of the same shape which turned out harder than expected. Intricate parts of the template were hard to create because the carbon fiber just fell apart when there wasn’t enough material to hold it together. Sharper scissors or using pre-preg carbon would have helped. 

I prepped the shoe with masking tape to protect the upper and started epoxying the carbon fiber to the shoe. Overall it was an easy process. I wrapped the shoe in peel-ply and saran wrap and then let it sit outside for a day.

The results were pretty good! I did notice some areas where the carbon was more bare. I think my distribution of epoxy wasn’t too even as I tried not to add too much epoxy. It felt pretty stiff even with 3 layers of carbon fiber. What surprised me was how sharp the carbon fiber became after the epoxy. I was pricked so many times that I wonder if it was a safety hazard. Carbon fiber splinters are probably not good. I sanded down the shoe and tried to remove as much masking tape as I could. Looking back, I probably didn’t need the masking tape anyways. I drilled the 3 holes again for the cleat and added some skateboarding tape for grip and we were off to try it out!

The Review

I bravely took the shoe out for a shop ride on Sunday as the first ride. It was 50 miles but I also took my regular right shoe as backup. The new shoe clipped in and out easily and even made the nice snapping noise. It felt stable and secure even during the climbs. From a functionality point of view, they worked. From a comfort point of view it left a bit to be desired.

Here were some of the comfort problems. Because I used a soccer shoe, the toe box was smaller than most shoes; I guess it would be beneficial for a soccer player. This wasn’t great for cycling as it felt a little too snug after my foot warmed up. The Shimano cleat I used ever so slightly elevated the area around the ball of my foot but I didn’t offset that with something at the heel of my foot. This cause a hot spot problem which actually led me to swap back to my regular shoe. The last problem was a comfort and usability problem. Soccer cleats don’t have the best lateral stiffness so unclipping took more force than I would like. It wasn’t difficult; when the heel moved out-bound, it would unclip the shoe a little bit later than usual. As a result, it didn’t feel very ergonomic and was a little annoying.

What’s next?

Well, the shoe worked. Though not super well. I think using a soccer cleat as a donor shoe because it looked cool helped spark conversations but wasn’t the best donor shoe. I probably will not make the left pair with the soccer shoes and may find a new donor shoe. To be honest, my current cycling shoes are going to hit the bin soon so I might just go and purchase a new pair of cycling shoes.

Overall this was a very fun project. I was glad I got into carbon fiber and now have the materials on the ready for new projects. Maybe next time i will have something more permanent to share :)

[adventureswithedmund]

Garage Door Opener++

All garage door source code: https://github.com/edfungus/homebase

eevee, a HTTP/MQTT/Kafka bridge source code: https://github.com/edfungus/eevee

Well this look’s a little familiar doesn’t it? I have updated the garage door recently to include a bit more features like authentication. The garage door hardware is still running on the local network therefore requires the user to be within the area of the house. Here are some of the new features I added.

Facebook Authentication

Since all the roommates have Facebook (and who doesn’t these days?) I thought that a good way to authenticate the user is via Facebook login. Initially I let the ESP8266 handle the HTTP request; however, in the end I decided to use a Raspberry Pi instead. The ESP8266 was quite slow, adding about 1-2 second per garage toggling, and would sometimes crash. I spent a decently amount of type checking for memory leaks and tried to optimize memory usage but couldn’t get it to be stable. Some searching online for the error code showed that it might be power management related (?). Anyways, the ESP8266 handles only the garage control and the Raspberry Pi handles the Facebook authentication. Since Facebook used OAuth2, it was pretty simple to verify the user’s token and getting their user id to check whether or not they have permissions. 

Open and Close !?

In the past, the only command for the garage door is to toggle the garage because we didn’t know the state of the door at any given time. To implement open and close, we would need to know the garage’s given state in order to know whether or not any action needs to be taken. I added an sonar sensor to the ESP8266 which can tell whether the garage door was all the way open. The sonar sensor was more finicky that I expected and required a smooth flat surface to give a non-noisy output (hence why a piece of cardboard is stuck to the door). I stripped some old ethernet cable in order to get the long length of wire. A shortcoming from the location of the sonar is that the garage door will show “closed” even if it is partially open. I could have routed the sonar to the front of the garage where it will detect a sure close, but given the difficulty in routing the wire along side of our garage, I settled for this solution. 

Added Complexity with Rasberry Pi and MQTT (and points of failure) 

As explained above, the ESP8266 wasn’t great at authenticating the using via REST so I used a Raspberry Pi instead. This meant the request from the user would go through the frontend, to the Pi and then to the ESP8266. I wanted to move the ESP8266 off REST and to something more light weight like MQTT. So I used the popular Mosquitto MQTT broker on the Pi with authentication to facilitate the communication between the Pi and the ESP8266.

I recently wrote a protocol bridge library in Go (eevee) which routes HTTP, MQTT and Kafka messages based on corresponding, routes and topic names. Simply by using that library, I was able to route the HTTP traffic from a couple routes and forward the messages directly to the corresponding MQTT topics (to the ESP8266). Using a HTTP handler wrapper in Go, I added the Facebook authentication so that requests made to the Pi will be authenticated.

So in the end...

... the garage door works great! It is nice to see the status of the garage door while around the house and the Facebook authentication ensures that the right people get access. Granted, this is not a highly secure system at all. For example, SSL should be used on the broker and ESP8266 but I haven’t figured out the ESP8266 side yet. Also there is an insane amount of pieces that need to be up just to open the garage door. If any of the following goes down, I’ll be out of luck: Google Cloud Platform, Facebook Authentication, my Raspberry Pi (MQTT broker and eevee) and my ESP8266. Luckily everything has been pretty stable and reliable enough to leave the house without my keys!

[adventureswithedmund]

Chat app with CSS animations

live chat demo, interface only demo, github

Even though I have worked on frontend development for a while, I have yet to really do anything with CSS animations. So I decided to learn a bit about the transform and transition properties. By no means was it a deep dive and I learn the best by doing, so I made this chat interface...

The Interface and CSS 

Adding the CSS animations was easy but managing the timing took some tweaking. Most of the animation resides in the introduction and some per message. It is tempting to add many animations since it is so easy, but that could easily overwhelm the experience. 

With that said, having that extra fade or slide can change the feel of the application drastically. When the animations were slow or the time gap between animations were too long, the app felt slow and unresponsive. While having too fast of animations would feel like sensory overload. Finding the balance is key.

Adding the Chat Functionality

An interface is cool, but what about the chat feature? At GE I work on the Event Hub team who creates a publish/subscribe messaging service. So I decided to use Event Hub and the chat interface to make a chat client. I added some tweaks to the interface to display username and commands to show online users. 

Extra Tidbits

The interface is fairly flexible on screen size which means in works well on mobile too!

From a UX perspective, I wanted it to be very minimal yet understandable 

The “/commands” is easily extensible for any keyword 

[adventureswithedmund]

Phone + ESP8266 garage door opener

I share a house with several other fellow coworkers to keep the cost low. The Bay area is quite expensive. We all have cars but the garage only has two spots and two garage door opener. So I decided to just hack up a simple device to open our garage door. It is quite simple and we can open the garage door with a press of a button!

Our garage door was a simple Liftmaster one that had a single push button to open/stop/close the garage. The easiest way to open the garage door is to trigger that button. I chose to do it electrically with a relay. With some soldering work, the wires were ready to be attached to the relay and the arduino. 

So I didn’t actually use an arduino but rather an ESP-8266 but conceptually it is similar. The ESP-8266 already has wifi built in which made it easy to make a landing webpage. The relay triggers the button as if I pushed the button; however, the relay timing took several tries to get consistent. The ESP-8266 is powered by a wall micro usb charger and connects to our home router. 

To actually trigger the garage, you would just visit the website hosted on the ESP-8266 it self and hit the “toggle gate” button. Unfortunately, like the physical button, it cannot tell you the current state of the garage. I used Lua script and some simple Bootstrap CSS to make a pretty. Going with the theme of the house, I also add a Twitter feed to keep things fresh. 

Overall it works quite well and we are all please with it. We longer have to take the keys to go running and whenever someone wants to park in the garage, they can always open the garage door without worrying whether or not they have the remote. The code is fairly simple so I will probably not push it to Github but I will provide upon request. 

[adventureswithedmund]

Navigate Google Maps on computer by tilting your phone

What if I told you any phone can control Google Maps on any computer, anywhere! (Of course you need internet and what not) I made Map Flyer a while back, but forgot to post about it. 

Website, Github

I primarily wanted to take advantage of MQTT’s real-time and multidirectional messaging.

The phone and computer are logical devices to connect together because they are common devices people have. Map Flyer takes advantage of the connection to allow one device to control the map while the other device displays the map.

Using one device to control Google Maps while another displays it is a simple demonstration but useful in showing the power of real-time multidirectional messaging. Go to Map Flyer on your phone and computer and enter the same name for each device. Then when your phone is tilting, the map will move in that direction. If your phone is flat on a table, the map shouldn’t move. There are also zoom buttons too! It is pretty fun navigating from one place to another in the world.

Some more details about Map Flyer. It uses of the device’s hardware accelerometer to provide a physical experience in navigating the map. A dynamic connection between any two devices is generate by using the given name in the creation of making a MQTT topic. There was also some math that went into correctly scaling the size of movement based on the zoom level.

[adventureswithedmund]

Human Brains

Before I go to bed, I was thinking that humans do not really have memory. We do not have "memory" which is analogous to flash drives. We do not actually store copies of sound or images in our head. I propose that everything we "know" or "remember" are just complex resolutions of patterns. 

If one looks at a movie poster or flyer, I bet one will not remember the whole poster. The argument can be made that the parts of the poster we didn't remember were not inspected carefully before hand. But before looking again at the picture, it is very likely that one can imagine what could have been on the poster; filling the gaps in our "memory". In fact, one can probably dream up of another very compelling movie poster that quite frankly never existed. Like memory, imagination is just more pattern recognition and generation. Therefore human thought and intelligence is really just complex pattern matching that has grown more and more complicated by combining data that comes through all of our senses. We also posses the ability to augment and rearrange these patterns in our brain.  

Pattern matching allows prediction which is more valuable when it is more accurate. As a result, it compels natural selection to commend those who are able to predict better or have more powerful pattern engines. 

With that said,  any pattern that our brain cannot match or discover would be the things that our brains and us can never understand

Patterns are difficult to find and difficult to extrapolate. For good machine learning, the program must do both well and find ways to understandably store the data via patterns.

[adventureswithedmund]

Low Latency Screen Capture Streaming to Play AimBooster

Github: https://github.com/edfungus/MQTT-Screen-Streamer

In my last post, I was trying to do text detection of my computer screen which was streamed over from my Windows computer to an Ubuntu VM on the same machine. The latency was a minimum of 1 second over RTMP (OBS client and NGINX server) even with 10 fps. This was not really acceptable when you want your computer to use computer vision to play AimBooster.

I wanted to use OpenCV to playing a game called AimBooster where one tries to click on the dots when they appear on the screen before the dot disappears. This game is decently difficult with a mouse, but a computer should be able to nail it everytime.

RTMP to MQTT

RTMP streaming was too slow so I went and wrote a simple streamer that took screen captures, encode to base64 and send them over MQTT. Some caveats:

I didn’t need any audio

I didn’t need frames to come in smoothly or evenly, they just had to come in a timely fashion

I wrote up the capture client and the reader client. The capture client only runs on Windows because it uses the win32gui library and is pretty fast. It also is in charge of performing the mouse clicks. The reader client analyzes the screen capture frame for the dot and sends back the x,y coordinates of the dot to the capture client.

Brokers

The broker I tried at first was mosquitto which is the broker used in Crouton. But it lagged out after ~5 minutes of running. The frames would get really backlogged and the frames being analyzed will be very outdated and out of sync with the game. Therefore I went to another broker which I used for testing of Crouton emqttd. It is written in Erlang, really fast and can handle 30 fps super well.

Finally

Overall the streaming solution was so good that I could watch videos via the stream and the audio would be only some what noticeably off, something you would have to look for. Also it played AimBooster pretty well too haha

[adventureswithedmund]

Text Detection

Github: https://github.com/edfungus/Text-Detection-from-RTMP-Stream

This is a brief write up on some OpenCV that I did to detect text on a streaming video of my computer screen. The setup includes running an Ubuntu vm on my Windows machine. The Windows machine used OBS to stream via RTMP to the Ubunut vm’s nginx RTMP server. The latency for such setup is around at best 1 sec (even when network is set to host-only to remove the router from the picture)

Basic Principles 

The basic principle to finding text is to turn the picture into a binary image where the areas with text stand out enough to be able to thresholded. I took two approaches to making the text stand out which are in the functions gradientText and cannyText in analyzer.py. 

The gradientText function uses a gradient morphological filter which is just a dilate operation followed by an erode operation. The cannyText function uses a canny filter which is an edge detection filter. They both try to make the text stick out.

I then connect these areas horizontally using another filter because words are generally horizontal in structure. I also use other parameters with width, height and size to get better accuracy on the text detection.

Making it Better

The text detection with the above idea works ok, but it picked up some other areas (mostly parts of pictures) which matched the filter but were not words. I tried to come up with other methods of improve the detection algorithm above. The first three functions in processor.py tries to improve text detection on the screen.

My first thought was to use the relative text locations to determine whether or not the text detected was likely to be text. I assumed that text which aligned vertically and horizontally would more likely be text. The screenshot below highlights red when text is strongly horizontally matching while green for strongly vertically matching.

Using this same concept, I also assumed that text closer to each other is more likely text and less likely an anomaly. This method was ok, but did not seem to give better performance compared to not using it at all.

Another simple check I did was to ensure that the width and height ratio resembled that of text. For example, text most likely will have a longer width than height. 

Each of these functions output a confidence score for each of the locations of interest. It was also written so that these scores can be combined to generate a final combined to score to determine whether this location is likely to be text or not. (combineMultiple function in main.py)

Latency Problems 

Using RTMP was an easy solution but was not very good. Latency was at least 1 second and sometimes up to 5 when processing. One thing I tried was to first find all the areas in the frame which has changed from the previous and only process the areas that have changed. This actually improved performance a lot. 

The rest of the functions in processor.py and simpleDifference in analyzer.py are dedicated to find the difference areas in the image, removing “found text” from regions touching the difference areas, finding text only in the difference areas and recombining the found text region in the difference areas with the previously found text. 

Here are some cool videos of finding areas that have changed on the screen:

Box The Text!

In in the drawer.py file I wrote functions that aid in drawing the boxes. It just allows to control thickness and color from and array of rectangles given from a text finding algorithm. 

Wrap Up

Overall this was a cool experiment. RTMP was not the best choice for this application because I didn’t really need to stream video since video includes both audio and correct frame timing (which introduces buffering). I also want to look for a low level method of grabbing frames and just sending them over to process. The reason i want to send frames to a vm to process is because in the future, I may want to stream the screen of my personal computer to a more powerful desktop for processing (and machine learning... :O) 

[adventureswithedmund]

Wrapping Up Crouton

Demo site: http://crouton.mybluemix.net/

Github: https://github.com/edfungus/Crouton

Just wanted to make a couple closure points about the projects I have been doing. Crouton is now finally an open source project that can be found on Github. There has been a good response from the community about Crouton. I’m glad people found it, wrote blog posts about it and even interacted on Github with stars, issues and pull requests. 

It was a really fun project to do especially when I wanted to make this solution with very limited infrastructure resources. Shout out to IBM Bluemix for hosting for free (and you can too!) 

I am probably not going to continue the development of this project unless I see a big need for it. I am going to be moving on to cooler things, no worries :)

[adventureswithedmund]

RGB Led ESP8266 with Crouton

Crouton has been released for a while now so it is only logical to make some physical hardware devices for it! The first device I choose to make is a simple RGB Led coupled with the RGB dashboard card on Crouton. 

Crouton: http://crouton.mybluemix.net/

Crouton Github: https://github.com/edfungus/Crouton

Things you will need:

ESP8266-12 (or anything but we will use this one for this guide)

RGB Common Anode (common cathode will work too)

3 x 220 ohm resistors 

1 x LM1117t

FTDI cable

ESPtool.py, Luatool.py, NodeMCU firmare (will have links below)

Wires, cutters, soldering iron, etc ..

Getting started:

First we will setup the ESP8266 in terms of getting power and installing NodeMCU firmware. If you have down all or some of that, feel free to skip ahead.

First using the LM1117t, we will convert the 5v from the USB FTDI connection to 3.3v for the ESP8266. Wire the 5v from the FTDI to the input of the LM1117t. Also wire GND to GND. 

With the FTDI connected to the LM1117t, lets connect the 3.3v power and the FTDI serial to the ESP8266. Use the diagram above to wire the ESP8266 correctly for programming mode. This is the mode used to reprogram the firmware on the ESP8266.

esptool.py write_flash --port COM_PORT_HERE 0x00000 firmware.bin

Download ESPtool.py from Github and the NodeMCU firmware from Github. The integer version will do fine. First plugin your FTDI cable to the computer and figure out with com port it is using. Then using the following command but with the correct com port and name of the firmware, flash the ESP266. It should saying “Erasing Flash”, then “Writing”.

After that is done flashing, rewire the ESP8266 like above notably unplugging a couple wires. Rebooting the ESP8266 in this mode is for normal operations. In this case, it will be running the NodeMCU “operating system” which can interpret Lua script files. The ESP8266 will always run the init.lua file upon booting. 

Transferring Lua script files:

Head over to the Crouton Github for the lua scripts needed. These are the files with code that makes the ESP8266 actually connect with Crouton and control the RGB Led. Look inside the “lua” folder for the init.lua file and the rgb-new.lua file.

python luatool.py --port COM_PORT_HERE --src init.lua --dest init.lua

Go into the init.lua file and enter your wifi details, overwriting mine. (whoops now everyone knows my wifi info :O )Then transfer over the init.lua file with the command above.

python luatool.py --port COM_PORT_HERE --src rgb-new.lua --dest main.lua

Then the rgb-new.lua file as main.lua.

Connecting RBG Led:

Now we want to connect The RGB Led to our ESP8266. The longest leg connects to 3.3v because it is a common anode RGB. A common cathode RGB will connect to GND instead. Now from each of the three legs will need to connect them to the ESP8266. Remember to add the 22o ohm resistor to each line between the led and the ESP8266 to prevent bad things from happening. The Lua script is programmed to the following RGB to pin combination.

Red -> pin 4

Green -> pin 14

Blue -> pin 12

Using Crouton:

Now comes the fun part! Hook up your ESP8266 to your computer and open a serial monitor (like the one that comes with the Arduino editor) and when you reboot the device, it will tell you whether it connected to Wifi/MQTT Broker but more importantly, it will tell you the name of the device. You can also change the name to what ever you want in the code in rgb-new.lua (line 31 as of now) 

On Crouton, head over the Connections page and use the default MQTT Broker. You can change the broker in rgb-new.lua if you want to use a different one but your ESP8266 and Crouton must be on the same broker.

Then enter the esp8266′s device name you got form the serial output or what ever you changed it to be.

If it connected successfully, then go to Dashboards and start playing with eh RGB Led! If you could not connect, make sure the ESP8266 is actually connecting to your Wifi and that it is connecting also to the broker. 

I hope you all enjoyed this guide! If you have any questions feel free to shoot me an email: [email protected] There are many other dashboard cards you can use in Crouton so be creative and connect up your IOT devices!

Twitter: @edfungus

Watch me code Crouton on Livecoding.tv!

[adventureswithedmund]

Crouton

Live demo

Crouton is a dashboard that lets you visualize and control your IOT devices with minimal setup. Essentially, it is the easiest dashboard to setup for any IOT hardware enthusiast using only MQTT and JSON.

Crouton is here! It is a dashboard for your IOT devices! Without any configuration on Crouton, upon connecting your device, Crouton will know exactly what dashboard elements to sure you for your devices. You can see and interact with your IOT devices with minimal setup!

This is done by sending a JSON from your IOT device via a MQTT network. If your devices do not already run on a MQTT network, it better! MQTT is a protocol made for IOT devices and it uses a much better publish and subscribe model. 

All you need for Crouton is your DIY devices and a MQTT network (free ones online!). Also documentation to help you get connected. And github if you want to run Crouton on your own server (or RaspberryPi!)

[adventureswithedmund]

Salads and Croutons!

[Update! Some things have changed. Check out this post]

A crouton is an internet of things node that can be anything from a light switch to your thermostat. A salad connects all the croutons together so that you can view and control all your croutons. Dressings are analytics that you can perform on the data collected.

Welovesalads.tumblr.com - My new project aims to connect the ‘internet of things’. This site is test bed for the development of a MQTT Client dashboard to show and control all your attached MQTT IOT devices. Pretty much a plug and play IOT devices system in the end. Development with MQTT, Angular.js and ESP8266 Wifi microcontroller

MQTT is the protocol I have chose for this project. It is lightweight and perfect for devices that talk intermittently in a possibly unstable connection. Using MQTT.js and emqttd but later Mosca + Redis

Angular.js is a get client side templating and more module. The two-way binding is amazing. Using MQTT.js also to talk to IOT devices.

Cloud Foundry is a crazy awesome tool which separates the infrastructure and platform from your service. I can deploy a Node.js server in Cloud Foundry without knowing the computer or operating system is on. It can be hosted on Amazon or IBM or somewhere else and nothing changes. Currently I am on Tumblr but will be running the site on Cloud Foundry with IBM Bluemix’s free tier along with the MQTT broker (Mosca .. also node based) and Postgresql database for timeseries data.

ESP8266 is the basis for our hardware. It is super cheap, has wifi and also an easily programmable microcontroller. Currently I am using Lua scripting because I had some issues with native C interrupts.

My current focus with this project is on the web interface and UX that allows the management of different ESP8266 IOTs including connection, control and display. I want this system to be generalized and expandable. After that I will focus on the hardware side of the project and finally some analysis of the time series data.

[adventureswithedmund]

The 40hr week life

Now that I am out of the University of Texas with a Biomedical Engineering degree what would I do? I didn’t quite expect to work for GE and much less did I expect to work as a software engineer. But I guess my programming projects and my drive to learn and make “things” payed off in a way. 

All is well; I am currently in California at GE in a leadership program called “Software Development Leadership Program”. It is a new program but based on GE’s other famous leadership programs. We just finished the 2 week orientation which included classes on Agile, etiquette, meeting facilitation, design thinking and many other things. Now I am just finishing up my week business trip to Atlanta, Virginia and Washington D.C. meeting the rest of my team. There are even people from India in team. I have learned a lot and has been really fun. Ready to get back to developing however.  

But enough of the boring things, what are the fun things I have been doing??

Cycling to work every day! I live super close to work. Of course I am still doing long rides as I can.

German on Duolingo is going well. ~12% fluent (I think that metric is optimistic)

And... a new project related to programming/tech/internet of things!

[adventureswithedmund]

Some Updates!

Many things have happened since I have last posted. I crashed twice (yes, twice), got OSX installed on to my Dell 9530 XPS, and made some money from taking people’s graduation picture (and an engagement shoot too!)

First the hackintosh.. People always ask me why I do this and I tell them it is fun and challenging. It is pretty tough to get the whole machine working including sleep, power management and WiFi. Luckily, I found a guide for my laptop pretty much exactly and got mostly everything working! I also went on eBay to grab a compatible WiFi card (Broadcom BCM94352Z) since my Intel one doesn’t work. It was pretty hard finding a card since my Dell uses the new NGFF M.2 shape. I am using this as my main machine now. Python is perfect for my music server and will eventually be my server!

My crashes..well lets say they were not fun. Both time I had to purchase a new helmet (at like $70 :( ) but bike wasn’t broken. The first time I flipped because my front wheel got caught in a fence after we will gunning sharp turns. Injured my left shoulder but other things were fine. The second crash was a bit worst. I was in a race and someone slammed into me from the side (why? I don’t know) and we both flew to the side of the road at 24+ mph. I had lost skin from fingers and again landed on my left shoulder. The worst has yet to come. I thought I was fine for the most part until l realized I was peeing blood!!! We went to TWO hospitals after that and I had to pee in a bag for 3 weeks. What happened was that I got a urethral laceration so it could have been worst. It sucked a lot. I am back now and riding, all is good. 

Never expect this to happen. I decided to shoot graduation photos for some extra money for my biking stuff... and I got a couple gigs. But what was something more exciting was that I was asked to shoot an engagement shoot! Not only that, the couple flew in from KOREA! ..and they don’t know English.. and they wanted me to shoot in both San Antonio AND Austin. Wow. Well it went well and I was adequately compensated :) 

Lastly, I am graduating next week. College at UT has been fun. Will miss it and the people a bit but I am also excited to go to San Ramon to work at GE!

[adventureswithedmund]

Biking update

Yes that is right, that is a new 2015 Felt F1 frameset :D My Univega steel bike is still going well but I figure I would give myself a graduation present. I will building this bike with the same Sram Force groupset and Bontrager Race X Lite wheels. My Univega will get Sram Apex and serve as a commuter when I move to California!

But do not worry, I am definitely still training hard. I know the new Felt will probably not make me any faster. In the last couple of weeks, I entered two races. The Texas State collegiate race was hard because it was in 36 degree rainy weather and since it was the first time, I had zero expectations. The second race I entered was the Austin Driveway Series. Wow, what an amazing race! The race is held on a car race track so the roads were super smooth. The only problem was that the field is Cat 4/5 so it was also super fast! An average of 25.6 mph for 30 minutes according to my friend who got 5th. I sadly was dropped after lap 3 even though I was basically going all out. 

Training wise, I have been riding mid distance 33 mile rides. I just recently added running and intervals. Also I did a long ride to San Marcos and Texas State just recently! 69 miles at almost 17 mph. Not bad I would say :)

[adventureswithedmund]

Music Server

I'm back! This semester has been filled with cycling and photography but I have manage to get some programming in. Let me introduce you to the Music Server!

[Do you see the search bar??]

[Live search!!]

[When you click on an album or artist, you go to their page indicated by gray text rather than yellow. But it is still editable for live searching!!]

What is it? This software will allow you to control music playing on a server from anywhere! Wait, how is that different from all those commercial WiFi or bluetooth solutions? Well the music in this case is not streamed from your phone or computer. Instead, the music is stored on the server itself. This opens up some interesting opportunities.

Imagine at home, the main speakers are connected to your music server. Pulling up a webpage allows you to change the song being played. In real-time, someone else in the room can also queue music to be played! If you have a party, you can let your guests queue what ever music is on the server. Eventually you can queue Youtube, Spotify for even a larger access to music!

Why am I doing this? Well, it is some what useful. I don't have to be 'in range' with bluetooth or have a computer tethered to the speakers. It would be like a proper Hifi with internet control! But the real reason I am making this is because this a practice run of making a server that has outside (internet) communication that can control utilities inside a house. In the future, this program will be loaded onto a raspberrypi which also has IO to control hardware like lights, locks, cameras, etc. Home automation is coming! The server is the easiest and cheaper way to get started.

How am I doing this I really like Python so I choose Flask as a framework and it has been working quite well. Eventually I will move to nginx to handle the server request but right now I am just using the Flask debugging server. Ubuntu development environment. Dual booting is a little annoying, a mac would be awesome...

I love fancy interface designs! No not gaudy, but efficient and sometimes esoteric. The live-search is just so amazing when searching through music. Click once to queue and play, double click to queue. In the queue, one click to play and double click to remove. Damn simple.

Github I needed some sort of version control better than just copying the folder as backup haha. So there's the link for those who are interested.

It is definitely far from done. It can play music from queue in repeat right now, scan for music and even grab ID3 data. However, right now I am investigating websockets to update the 'now playing' page.