It's been a while since I first installed prosody on Agayon.be. I use it to experiments with my bots, to keep contact with the XMPP community and discover new cool stuffs to do.
Recently I struggled a bit because I wanted to hide the prosody small HTTP server behind my Proxy. For various reasons, I still use Apache 2.4 and I could not get it to work with prosody. I mostly use the HTTP server for bosh authentication with Converse.js and with the http_upload module.
When the 5281 port was accessible and Prosody handled the requests directly on the internet it worked well. But when I followed the documentation to use a proxy, it stopped working.
All my PUT requests got a 404 error. I tested my setup with Slixmpp and the http_upload example.
During these weird time of corona virus crisis, like many, I am working from home and cancelled all social activities. Apart from the distanciation, the virus, the lack of activities, it has been a quite fruitful period as I work a lot on the robot. This articles sums up the latest developments and achievements.
My robot, the Agayon has now kite a lot more features. Among which:
Video streaming nearly in real time. The stream can be viewed in a web browser. It works quite well and has been used across the internet. Some friends or family members have visited me with the help of the robot.
The only drawback is that I have to limit the stream to 10 frames per second.
The setup is based on mjpg-streamer and the custom layout is available here.
The Agayon can now be remotely controlled with
A PS4 controller (Bluetooth)
REST API to control the robot.
A web interface that uses the API (see screenshot below)
Lidar mapping: It can be triggered by the PS4 controller, the web interface or XMPP. I use a RPLidar A1M8.
Data is saved in a file that can be analyzed afterward. No real time data processing for now.
XMPP: migration from the deprecated Sleekxmpp library to the more modern one Slixmpp.
Small turn angles for remote control. The robot was sometime too much brutal during rotation. This version adds the support for 'gentle turns'. During the new gentle turns, only one wheel is moving. The 'normal turn' remains and it rotates both wheels in opposite direction.
Start and Stop Webcam streaming from socket. It relies on the restart_stream.sh and stop_stream.shscripts. It is used from the web interface and XMPP.
Mapping: simple scan and logging of all mapping data. It Saves a picture (polar graph) for each snapshot.
Arduino communication: serial data to get the ultrasonic sensors measurements.
The rover can be remotely controlled with a PS4 controller. Events are caught and instructions are sent through serial communication to Arduino.
I created a small Serial manager to handle orders from the RPI. It can be used to remotely control the motors, get ultrasonic measurements, change speed, change mode (incremental or directly to a mode number), capacity to turn a little (small angles but not that small )
Send formatted serial data that can be easily parsed by the RPI process (odometry).
A lot of bug fix
The web remote control is based on a small webservice depending on with the Flask framework. It is served with uWSGI with the help of Nginx. The sources are available here. It is used to control the robot with a web interface but any client able to use a web API can use it. Maybe an Android client will follow?
When the video streaming service is unavailable, a 503 error is displayed with a custom page.
I use it to launch the Webcam streaming directly from the navigator.
All the configuration, HTML pages, scripts and config files are available in a separate repository.
Once mjpg_streamer is launched, it serves a small page to interact with the API.
This picture was taken at the beginning of a ~4m narrow hallway.
Here are some pictures token in my apartment. As long as no Lidar measurment has been made, a cartoonish picture of the robot is displayed.
As the robot is going forward, the mapping is updated by clicking on the Lidar camera button. For unknown reason, the left/right mapping is inverted in the picture.
There are still a lot of room for improvements. Here are some ideas to occupy me in the future:
Automatically save a camera picture when the lidar mapping is triggered.
Detect the kernel messages about battery. When the voltage is too low, the information is logged by the kernel and can be found with systemd. The idea would be to shut down the robot when necessary to avoid SD card memory corruption.
Apply some nice OpenCV filter to obtain a transformed video stream line in the movie Terminator. :-)
Use one button to trigger the video recording. Use a blinking LED to let known that it is "On Air".
Recording movies with the camera
Recording a movie with OpenCV can be done in less than 25 python lines. It works well with my old Logitech C170 but for some reasons it did not work out of the box with my Microsoft LifeCam Studio.
Dear visitor, if you have such a camera, here the magic trick to produce your new feature movie ! First, make sure to have this model with the lsudb command:
Bus 001 Device 007: ID 045e:0811 Microsoft Corp. Microsoft® LifeCam Studio(TM)
The camera produce naturally a MJPG stream. I would sum it up a stream of JPEG pictures. By default, the VideoWriter class will produce empty video files. Unfortunately, by default there are no debug message to help you.
Don't forget to use the following environment variables during your hacking sessions:
To make it work, you need to tune the camera settings before obtaining the desired result. The complete program is displayed here. Follow the comments to see where the magic happens.
importcv2importoscap=cv2.VideoCapture(0)ifnotcap.isOpened():raiseIOError("Cannot open webcam")# Magic number corresponding to a MJPG Streamcodec=0x47504A4D# You actually says that your camera produces such a streamcap.set(cv2.CAP_PROP_FOURCC,codec)# We fix the resolution, the framerate and use exactly the same in the VideoWriter arguments cap.set(cv2.CAP_PROP_FRAME_WIDTH,640)cap.set(cv2.CAP_PROP_FRAME_HEIGHT,480)# The framerate is not limitating herecap.set(cv2.CAP_PROP_FPS,20.0)# next we define the filename, the writer options and start the infinite loop.videoname=os.path.join('/tmp/',f"output.avi")fourcc=cv2.VideoWriter_fourcc(*'MJPG')video_writer=cv2.VideoWriter(videoname,fourcc,20.0,(640,480))whileTrue:ret,frame=cap.read()video_writer.write(frame)# When the Q key is pressed, the loop is stoppedifcv2.waitKey(1)&0xFF==27:breakcap.release()video_writer.release()cv2.destroyAllWindows()
You can find here the webserver and uWSGI configurations used to make it work.
During the Christmas holidays, I took the time to work on the Agayon. I hope that the mechanical parts are almost finished and I will be able to focus on the code in the following weeks/months.
Do as you’re told
A few years ago, I have been gifted with a PS4 controller to play on my Retropie Setup. These are quality controllers but Sony does not like makers. My model cannot be used with vanilla bluetooth drivers on my Pie. And for unknown reasons, the alternative driver ds4drv does not works for me. I purchased a 5m USB mini cable to not disappoint my lovely niece and nephew. They can plan with Supertuxkart for hours. But it's not the subject of this article. I plan to use the Agayon not only in my apartment but also to go on tour with it ! (let's dream a bit). Unfortunately, it is already quite heavy and it could be cool to drive it from my door to the car. I could dream again and imagine a system where it recognize me after a little bit of training like a dog or a case. It seems possible but why not use the controller at first. It should be easier and quick to implement. Moreover, the kids and friends seems to love the idea. So let's do it ! I will keep you updated when it is reliable and easy to use.
I have finally mounted the Lidar on the Agayon. The Raspberry Pi 2 was too slow to handle the data but the Pi 4 does well the job.
My living room
The following animation has been made with the animate.py script from the RPLIDAR repository.
Last year I discovered the Raznot project and it inspired me. According to the readme, the RazTot is an easy DIY project which allows you to remotely control a roving security camera securely from your browser. After some tests, I decided to not use this project because the flask server and interface would not so nicely integrate with my R1D3 base code. I only use Janus Janus, a general purpose WebRTC server to stream from the robot in a generic web page for now. Unfortunately, my Microsoft LifeCam Studio does not produce a stream compatible with Janus. I need to transcode the MPEG-4 video stream to H264 with ffmpeg to see it in a browser. I hope to be able to release the code in the following weeks.
After months of work on the Agayon, I can present some significant improvements ! This article is a little bit longer than the previous ones but it worth the read!
Software updates !
During the past few weeks, the code base of the Agayon has been updated. I forked my own project, r1d2 to update it. The new repository is named r1d3. I hesitated a long time before forking it. As the hardware base of the Agayon completely changed, I preferred to change the code name to maintain coherence between hardware and software.
As I made tests with SLeekXMPP to control the bot, I observed some problems with Gajim. The Ad-Hoc extension allows one to send commands to an XMPP bot. R1D3 displays the following menus and submenus (in french):
When I try to use the "execute" button, SleekXMPP start a new session and Gajim complains that the session identifier has changed. I reported the problem to SleekXMPP and its fork SliXMPP. The XMPP community is great and Maxime Buquet responded quickly. To quote him, there are two problems (see the bug report for the whole explanations):
Slixmpp shouldn't assume execute is the start of a command
I don't see a place in the XEP that says that next or execute can be equivalent to complete. What to do?
The serial interface of the SHT7x is optimized for sensor readout and effective power consumption. The sensor cannot be addressed by I2C protocol, however, the sensor can be connected to an I2C bus without interference with other devices connected to the bus. Microcontroller must switch between protocols.
One ground to rule them all
I have been advised to use an epoxy base coated with a copper layer. The aim is to connect it to the negative pole of the battery. It is really useful because it decrease the wiring. The perfboards are fixed on metallics spacer bars to avoid shortcuts.
I²C is a bus communication that allows multiple device to communicate with each other.
I2C devices are recognized by the Arduino (5V) and the Raspberry PI (3.3V) with the help of a level shifter.
Scanning...I2C device found at address 0x1D !I2C device found at address 0x20 !I2C device found at address 0x6B !I2C device found at address 0x70 !done
user$ i2cdetect -y 1 0 1 2 3 4 5 6 7 8 9 a b c d e f00: --------------------------10: -------------------------- 1d ----20: 20 ------------------------------30: --------------------------------40: --------------------------------50: --------------------------------60: ---------------------- 6b --------70: 70 --------------
"Scaffolding" with hot glue
During the past few months, my best friend has been my hot glue gun. I was skeptical at first but really much effective and fun. I used it to insulate some connectors. In Liège, we would say "mettre une noquette de colle" which translates to "put a knob of glue".
Evolution of the frame
A : Battery
B : Level shifter between Arduino (5V) and Raspberry Pi (3.3V)
C : Arduino Mega
D : Power lines and I2C (12V, 5 V, 3.3V, SDA 5V, SCL 5V, SDA 3.3V, SCL 3.3V)
E : Raspberry Pi (in his case)
F : Buttons and their pull down (3.3V or 5V depending on the GPIO)
G : LEDs
New (old) Oscilloscope
One of my colleague has been cleaning his lab, and he asked me if I was interested to have an old 20 MHz oscilloscope. I gladly accepted. It is a 34 years old Circuitmate 9020 (bought in 1985).
I will use it for I2C debugging and visualization.
The hardware is almost done. I am happy to have a nice reliable base. I hope to be able to drive it with my smartphone soon. I will continue the programming to add the mapping functionality and a nice demo mode.
Running a server on the internet is quite an adventure but it's not always easy to keep track of log files, security and potential threads. This article describes some actions carried out on this server to mitigate script kiddies.
Odoo 12.0 is out since october. I am currently investigating the differences with previous versions to update the instance of the association Les Compagnons du CEP. A lot of changes have been made in a few years but the workflow stays about the same. This article describes my workflow, the backup policy, how a module was used and fixed to restore a missing feature. Finally, the changes in my custom product import function are presented.