Winter 2026 MSE403H1 Practical OT2 Liquid colour matching demo

[kelvinchow23]

Continuing from last year's course #152

[kelvinchow23]

The previous 3D printed sensor housing likely wore out over time causing an issue where the pipette would pick up the sensor and then drop it immediately. This was temporarily fixed by adding some super glue back to the interior of the fake tip pickup area to add some more material back. There's a backup sensor printed just in case, although the fake tip region looks different than the current one. This has been going for about 100-200 pickups now, haven't seen an issue yet

[kelvinchow23]

The wireless charger base doesn't seem well secure, the charging coil seems to dislodge frequently. My guess is when the sensor drops into the charging base, it shakes the base and dislodges the coil over time.
I also superglued some things together on the interior. The base could probably be designed better to securely house the coil in the future

[kelvinchow23]

Adding a comment from Jae, maybe black out the box to stabilize external light noise.
Seems like Sterling thought about adding a light here #152 (comment)

[kelvinchow23]

I've had to home the gantry twice today during runs. What i think is happening is when the pipette drops the tip back into the tip rack, it doesn't always drop the tip back into the right place. See image below, it hangs above the rack. The next time the OT2 tries to pickup from this position, it moves the gantry out of position and causes any future attempts to fail. Need to look into it further

[kelvinchow23]

continuing with the theme of immediate hack fixes, i filed the sharp edges down to avoid it getting caught on the lip. i wonder if this is a common issue, or if we're just picking up the same 3 tips so much they get deformed

[kelvinchow23]

why do i feel like the red isnt doing anything?

[sgbaird]

Hey, Kelvin!

The previous 3D printed sensor housing likely wore out over time causing an issue where the pipette would pick up the sensor and then drop it immediately. This was temporarily fixed by adding some super glue back to the interior of the fake tip pickup area to add some more material back. There's a backup sensor printed just in case, although the fake tip region looks different than the current one. This has been going for about 100-200 pickups now, haven't seen an issue yet

Hmm... Related to #64 (comment). Not sure which version you have printed (if it's out of date on https://github.com/AccelerationConsortium/wireless-color-sensor, lmk so I can fix)

In particular: "New design with a thickness of 2.5mm survived after around 7000 times stress-test. I updated the model file and replaced the LCM-demo with the new design."

See also #87 (comment) since changes to the design (and even reprinting the housing) can change the color sensor results.

The wireless charger base doesn't seem well secure, the charging coil seems to dislodge frequently. My guess is when the sensor drops into the charging base, it shakes the base and dislodges the coil over time.
I also superglued some things together on the interior. The base could probably be designed better to securely house the coil in the future

Good points. This was likely due to my overzealousness in asking for a low-hardware approach when Yanghuang was designing it (i.e., minimize number of items on bill of materials like nuts and screws and required tools).

Feel free to redesign or suggest improvements. If so, I recommend looking at #64 if you haven't already to get a sense of the discussion around prior designs.

Adding a comment from Jae, maybe black out the box to stabilize external light noise. Seems like Sterling thought about adding a light here #152 (comment)

Like you're mentioning, throwing a blackout curtain from amazon or a home depot or somewhere could help. I think the light below should be around somewhere still. Just left it always on and attached with tape to the top.

continuing with the theme of immediate hack fixes, i filed the sharp edges down to avoid it getting caught on the lip. i wonder if this is a common issue, or if we're just picking up the same 3 tips so much they get deformed

Maybe toss those ones and swap them out with tips from other locations? Based on a hazy memory, Opentrons doesn't recommend reusing tips for that reason I think.

why do i feel like the red isnt doing anything?

Yeah, looks like something's up if you're doing a random assortment of mixtures. I'm sure you're troubleshooting, lmk if you want me to comment

Thank you for keeping this alive.

[kelvinchow23]

Thanks Sterling! Hope you're doing well! Do you know if there are notes or a general protocol how to set this up? There's a few things I'm not sure how involved Yanghuang was while the experiments were running to keep it functional. For instance, did th pipette tips need to replaced frequently, how often does the OT2 need to be homed, if there is a concentration for the stock colour solutions, etc.

[kelvinchow23]

the sensor went down, no idea what's happened, need to investigate after. The OT2 got stuck over the wellplate, which I've never seen before. Usually when there is a loss in communication of the sensor, there is a timeout and the sensor OT2 will return the sensor back to the housing and return home. weird.

regardless, the immediate "fix" was to keep the sensor wired. seems to only like my laptop and not just power. guess i'm stuck here

[sgbaird]

Might need to pull in @Neil-YL a bit to comment if he's available (no worries if not).

Do you know if there are notes or a general protocol how to set this up?

Primarily it's scattered throughout the issues. There wasn't a single handbook or guide, and we were having to be pretty dynamic in handling things as we were developing it. More of an excuse on my side.

There's a few things I'm not sure how involved Yanghuang was while the experiments were running to keep it functional. For instance, did th pipette tips need to replaced frequently, how often does the OT2 need to be homed, if there is a concentration for the stock colour solutions, etc.

Yanghuang was solely responsible for keeping it running and for most of the development (OT-2, wireless color sensor, livestreaming), so he's really the go-to. I would step in very occasionally, but we would discuss frequently (primarily over GitHub). When Yanghuang left, those tasks fell to me and I would ping him occasionally if I ran into something that I couldn't track down.

did the pipette tips need to replaced frequently

I think I had only done this once on my side, which I think was only because I had swapped red and yellow at some point and didn't want to cause confusion. I'm not sure if Yanghuang ever did. I don't remember having a conversation around that. Occasionally, I've noticed the behavior you mentioned with the tip not going back in all the way, but the pipette always seemed to be able to recover from it (i.e., just pick it up as normal).

how often does the OT2 need to be homed

I have a vague memory of Yanghuang mentioning something related to this.. infrequently I thought? If there was an error though, the "catch" statement would be to home the OT-2 (see e.g., #349 (comment)).

if there is a concentration for the stock colour solutions, etc.

I thought this might have been labeled on the stock containers.. they are these larger jars / capped flasks. Do you have those around? More on that in this thread: #152 (comment)

---

Mildly related, we had talked about tracking restocking steps: #434. Lower priority / overkill at this stage, except for proof-of-concept.

Completely random aside: #176, related to use of near-infrared measurement.

Sorry for the scattered thoughts.

---

For communication issues, I'd recommend simultaneously monitoring the Hugging Face logs, the HiveMQ logs, and the raw microcontroller logs (sounds like that's working for you if it's plugged into your computer and being run via Thonny though). I would have thought plugging in a power cable directly (so it's not running on battery power) would be the same as plugging into laptop.. Are you running the program on Thonny?

Happy to hop on a call

[sgbaird]

Another thing to keep in mind, if the tiny green LED onboard the Pico W is blinking every so often, that means it's "waiting for experiment requests" and should be fine. There are a couple buttons with the peripherals - one is reset I think, one is to turn off the battery power.

If it's something related to the sensor package, reading through the docs in https://accelerationconsortium.github.io/wireless-color-sensor/ might help (more from a mechanical design and setup point of view).

I mentioned to Gurpaul, but I also highly recommend switching back to the mqtt_as.py version instead of the mqtt.simple version if that hasn't been done already: https://github.com/AccelerationConsortium/wireless-color-sensor/blob/main/sensor_file/main.py

I tried mqtt.simple to try to deal with spotty wifi, but it was a bad idea. See peterhinch/micropython-mqtt#161 (comment). This was reverted back to mqtt_as.py in this PR. Now that it's no longer at NRC, shouldn't be an issue.

https://github.com/AccelerationConsortium/wireless-color-sensor/blob/main/sensor_file/test_sensor.ipynb can be used for testing the sensor. Same with https://huggingface.co/spaces/AccelerationConsortium/light-mixing. This can help short-circuit some of the debugging if you know the issue is with the Pico W package.

Another debugging step that can help is switching the network (e.g., to a phone mobile hotspot instead of the NOKIA)

[Neil-YL]

Hi guys. Good to see this demo is considered as one of the practical again. I would say it is a very rewarding moment when the students completed their practical and told you it is super cool.

Several comments on the discussion above:

1. General setup: power up OT2, power up sensor, SSH to OT2 (SSH used as trigger to initialise the code), wait until the machine home twice, send a test from HF. If all good, refill vials, replace tips and plate and refresh data on Mongodb.
2. Color mixture and tips: I premixed colors in three large jars before the course and refill the vials everyday morning and before I leave the office. You may stir the jars a bit before refilling the vials. I replaced the tips daily when the students were actively using the machine (for 2 wks?). I did witness several times tips did not fully drop into the tip rack but normally the machine will be able to pick up that tip. The reason I replace the tips daily is I noticed when dye dried out over night sometimes there were some clogging in the tip.
3. Sensor went down: I don't remember I had similar issue before but I would recommend to check the battery and the wireless charging pairs.
4. When OT2 should be homed: The only situation I remember I have to home the machine is when the code thought "I have a tip on pipette" but actually it doesn't. Example: sensor/tip return code does not execute for some reasons during the experiment. That is said if you see the machine stops running with something on the pipette and you want to manually dismount the things and recover the experience, you have to kill the py script and home (or power cycle) the machine. If you can run some codes to return things mounted on the pipette, you don't need to home. (but you still need to rerun the py script for the experience and it will home for you). If the experience run smoothly (for days), you don't need to manually home the machine.
5. Sensor timeout error: Iirc the sensor will be sent back to charging port if there is a timeout during measuring. Other than that the machine should function as a normal OT2. You can try to trigger timeout and debug by running the experiment with the sensor power off. I remembered once some issues on HF side stopped the experiment and left the sensor on the top of plate.

Ping if you need more info from me and I will try my best not to give you wrong answer based on my memory. All the best with this practical.

[kelvinchow23]

Appreciate the tips! Will need to take a closer look when there's downtime. I also don't have access to everything like the HF logs , which would probably be helpful. There were a few other issues.

1. The wrong data was being returned to the students. I noticed this too when I ran from the hugging face app. Last week, the experiment status was showing someone else's finished experiment instead of showing me the status of my current run. I don't know what the students were seeing on their end, but I could imagine that they are returning most of their own data, missing their last few experiments, which are replaced by the last few experiments of the previous student.
2. The youtube livestream playlist doesn't always show the latest video. This is very low priority right now as I assume very few students are looking at it. I assume it is a quick fix.
3. This one was fun. The pipette tip got stuck in the tip box and lifted the entire tip box. huh.

[Neil-YL]

1. This error had been captured before when students submit an experiment and close the HF page, this result will be sent to next student's output. I added a student number check but looks like this issue still preserve? May need to see if any method on HF can prevent this. I remembered I once had to send students' results on the DB since those are correctly recorded.
2. Iirc, this function is on HF. In late year's practical we manually update the live ID but later the HF has been updated to obtain the last (last, not latest) live video from the playlist. If the video id that HF got is not in correct playlist, you may need to adjust the playlist id. If it is confirmed the playlist is correct, then may need to look into the order of the video in the playlist. I remember we only sort a couple of video from the playlist but not all the video. (I could be wrong because this code has been updated several times). A simpler fix would be removed those video listed before the latest live one.
3. lol. May need to drain the leftover dye in the tip rack if it is intensively used over the weekend.

[kelvinchow23]

Another thing that we may need to look at is how the quota decrement is being handled. When a student submits an unsucssful job, it seems to count as an experiment. It should take away from succesful runs only. I'm going into mongodb and upping their quota for them to complete their runs