There was a good amount of progress that culminated this week!The uOttawa group got Tensorflow object detection working with the IR camera attached to the RPi! We were able to see what the objects were being detected as. This was just using the standard MobileNet model, and there were results such as surfboard in there too.
Beck completed the Bowie Brain Kit! It is now completely soldered. Next step is programming the first blink! As well, we had the chance to meet someone new, Queenie, who was interested in learning more about Robot Missions.
Hooray! It’s great to see it when progress meets a milestone. Congrats to the uOttawa team!
Meeting Queenie, who’s also interested in environmental robotics
Tensorflow is running!
Now setting up the RPi
We ran into an issue with this wifi, but switched to another and it worked
With the progress of the build guides coming along, today we took a pause to figure out how all of those pieces would fit together into a series. The result of this brainstorming session was seeing that there will likely need to be three series: a solo Bowie build (for makers building it on their own), a team Bowie build, and an educational group Bowie build. The edu build would be broken into phases, as seen. It looks simple, but there was a lot of erasing and re-writing while figuring this out on the whiteboards.
There was a bit of a problem being encountered after the Bowie rebuild. The XBees couldn’t communicate together. At first we were concerned this was going to be a hardware issue … that would be expensive parts. Alright, time to dive in to figuring out the bug.
First step was to see if any of the Xbees were receiving any data. Oh yeah, remember one of the previous issues with FTDI? This answer helped to fix the problem (from what we’ve seen, so far, until perhaps it will break again..).
Phew! What a relief! They were receiving data! Must be a code issue then. And wow, was it a code issue indeed. Meaning, there were sections of the code completely missing. The communications on that serial port were just vanished. What on earth! Well, anyway, we re-added it, and it was a huge relief to see the RSSI led from the XBee light up and to see Bowie move again, controlled by the Operator Interface.
Xbee comms are being seen through XCTU
Okay, the comms connection works, but no data is being seen.
This entire block of code was missing! How?! I’m fairly certain given this worked before that the code was there before!
Tonight the uOttawa group tested the IR camera. They connected it to the Raspberry Pi, and took pictures with it using raspistill. The test images were showing some objects with the overhead lights on and off. The IR leds that are beside the camera do a good job at illuminating the objects when the overhead lights are off. This progress showed that the IR images can be captured, and the next step is to pass them to Tensorflow for object detection. Beck added the headers to the Teensy 3.6 and inserted it in to the Bowie Brain.
Bowie has been disassembled since the summer when it was disassembled for the trip to France. With some of its pieces in different boxes, and sensors on different robots. It’s time to re-assemble a working Bowie again.
One of the pieces that broke was the front hinge, this was 3D printed and replaced.
The next piece that broke was a pad on the Teensy 3.6 lifted when trying to remove it. This is understandable, since the force of all the headers was much greater than that of a small copper footprint pour on some fibreglass. Alright, thanks to PJRC, we soldered new headers on and replaced this.
After that was fixing the Operator Interface on the green foam board Bowie. We ordered a replacement display, and all is displaying properly again.
There was a motor driver missing, but that was an easy find – it was in the prototype tree bot of course!
On the super bright lights board (version 1), some F headers are in place but the shrouding is missing. This was fixed by adding new F headers in place.
Alright, now it’s time to add everything back in place. All the electronics are in the main chassis now. Then, the air quality sensor on the back. Followed by the RPi on the top. We’re leaving the laser on the other side, because it’s still functional and cool. The claw is added too.
And now there is a working Bowie assembled again!
Added headers to new T3.6
This is what the alignment of the headers looked like
This is what broke, there should be 3 gold plated rectangles – not 2
New T3.6 installed
Comparing the lack of sand on the new one to the old one…
Display on the Operator Interface wasn’t working properly anymore
New one installed and working
It’s working again
Here’s a motor driver we were looking for
F header shrouding missing on the super bright lights board (v1.0)
Since on our system everything has been installed from various experiments and projects, we had to go back and be diligent in tracing the steps to install everything properly. The goal is to install all the software and Bowie libraries properly to get to a first blink. Today, the initial descriptions were written for this process. The next step will be to gather and make the thumbnail images for the steps. Then, running the .csv through the python script to publish as a page on the site.