Team E7: The Wand Makers

Carnegie Mellon ECE Capstone, Spring 2025 — Sharon Lai, Olina Zhang, Nadia Palar

Weekly Status Reports

Team Status Report for 4/26/2025
Posted by : hsuanjul Team Status Reports

Risks and Contingency Plans

  • The custom receiver is to be optimized. The current protocol we use is way too complicated to satisfy our needs. Should be done before the demo, if not we will demo the current working version.

Design Changes & Impact

There is no design changes at this point.

Schedule Updates

No major changes were made and the schedule remains as is.

Unit Tests and Analysis

IR transmission stability:

Sent cloned IR signals to LG TVs and Samsung TVs at different distances (1m, 2m, 3m, 4m, 5m) and angles (0, 15, 30, 45, 60, 75) to the TV. Successful transmission is 100% within 3m and 45 degrees. Successful transmission is 0% out of this range. The current transmission quality is not perfect but would need increase in voltage supply on PCB if we want to further improve it. At the same time, the drawback would not affect the demo, so we are not making changes to the design currently. We would need to address this problem if we want to sell it as a business product in the future.

CNN Model:

Confirmed that the model produces outputs with a success rate of 90% for five gestures and 80% for one gesture, based on performance evaluations from four different users after they learned how to perform each gesture and practice for 5 minutes. Verified that the loss function consistently produces finite values across different mini-batches. Monitored validation accuracy to evaluate model generalization.

No further changes were made to the model architecture, as the current version meets acceptable performance criteria. The CNN model was successfully developed, tested, and validated. The system is functioning as intended.

Progress / Photos

Wand controlling the TV in 1300 coves with cloned signal

Please see the project GitHub.

Sharon’s Status Report for 4/26/2025
Posted by : hsuanjul Sharon's Status Reports

Accomplishments

  • Validation for IR transmission: asked 3 different people to wave the wand at different angles and distances to the TV (cloned version). Note that the appliance being controlled also affects performance. Generally, LG TVs are more responsive than Samsung TVs. The transmission is stable within 3m and within 45 degrees to the normal of the receiver. Transmission is barely successful out of the range.
  • Printed a new wand case: the case was printed successfully but the button was lost, have contacted IDEATE to reprint.
  • Made purchase for final demo: order received by ECE, haven’t picked up

Schedule Update

I remain on schedule

Plans For Next Week

  • Pick up what we bought first thing Monday morning.
  • Make sure we have the button for the new wand case
  • Integrate custom receiver with the mechanical switch with Nadia.
  • Finish up with poster

Nadia’s Status Report for 4/26/2025
Posted by : npalar Nadia's Status Reports

Receiver is working more reliably, but there are still some slight firmware tweaks to try and optimize both the wand and custom receiver for demo day on Thursday. This involves performing some additional testing that will be reflected in our final report.

Next steps are doing the poster, video, and final report, as well as practicing for demo day.

Nadia’s Status Report for 4/19/2025
Posted by : npalar Nadia's Status Reports

This week, I continued expanding on the custom receiver functionality for the final demo as well as working on the final presentation slides. I also spent some time going over what to prepare since I am the presenter for this presentation.

I remain on progress. The final steps are to make sure that transmitting to the custom receiver is robust and to ensure that our final demo can go smoothly. We will also begin preparing the final report so that we have sufficient time to work on all of the deliverables due the week of the 28th.

Team Status Report for 4/19/2025
Posted by : hsuanjul Team Status Reports

Risks and Contingency Plans

  • Firmware Compatibility Issues : The updated firmware is compilable on different machines and we have fixed the previous bugs. However, functional issues may still occur after further testings. So far we have been able to fix the bugs, but if the plan doesn’t work, we would move back to the original firmware and build upon that.

Design Changes & Impact

There is no design changes at this point.

Schedule Updates

No major changes were made and the schedule remains as is.

Progress / Photos

Wand controlling the TV in 1300 coves with cloned signal

Please see the project GitHub.

Sharon’s Status Report for 4/19/2025
Posted by : hsuanjul Sharon's Status Reports

Accomplishments

  • Assisted Olina on CNN model: tested with various waving approaches of the gestures; increased dataset size to improve the inaccuracy problem during runtime
  • Assisted Nadia with the updated HAL firmware: debug, transmitter firmware is compilable, and is being flashed to the firmware

Schedule Update

I wanted to make some minor updates before printing the new wand case. So I am a little bit behind, but will be able to contact IDEATE on Monday.

Plans For Next Week

Print new wand case for final review.

Olina’s Status Report for 4/19/2025
Posted by : yunjiaz Olina's Status Reports

This week, I finalized the fix for the gesture recording issue we identified earlier.  I re-checked the dataset and re-identified the top six gestures with the top success rates. I modified the CNN model architecture accordingly.

Besides, I scaled up the dataset to twice the size, making the model even more robust. Consequently, the wand can correctly identify all the six target gestures with much-enhanced performance.

I remain on schedule as per the project timeline.

When developing, I tried out quite a number of model architectures with the aim of determining the one that best fits the task of classifying gestures based on the IMU. I started with a simple RNN, which could capture basic motion patterns but showed poor generalization. I then experimented with an LSTM architecture, which improved on training accuracy but overfitted quickly with our small dataset and required long training times. To balance spatial and temporal modeling, I chose Conv1D + LSTM, but it did not produce better accuracy. Lastly, I tested Conv1D-based CNN with two small-kernel convolutional layers, flatten, and dropout layers. This model had the highest validation accuracy throughout and was within the size limit.

I learned new skills in time-series modeling, model optimization, and data augmentation. I watched YouTube tutorials to learn key modeling techniques. I referred to TensorFlow documentation for implementing and tuning. Additionally, I read blog posts and Stack Overflow discussions to troubleshoot overfitting and understand best practices for dropout and kernel sizing. These helped me to quickly iterate and effectively tailor the model to meet our system constraints.

 

Nadia’s Status Report for 4/12/2025
Posted by : npalar Nadia's Status Reports

Since the last update, we were able to get the clone functionality working and successfully demo cloning and transmitting a LG TV signal during the interim demo. Additionally, I was able to make significant progress in finalizing the functionality for the custom receiver. This involved coming up with a custom IR signal to avoid accidentally aliasing existing signals.

I remain on schedule, though my workload may increase from needing to prepare a more polished and “real-world” demo for the custom receiver for the final presentation. Next steps are continuing to improve the custom receiver functionality and preparing for the final presentation.

Team Status Report for 4/12/2025
Posted by : hsuanjul Team Status Reports

Risks and Contingency Plans

  • Firmware Compatibility Issues : The updated firmware is compilable on different machines and we were able to successfully flash the firmware to the new PCB. However, functional issues may still occur after further testings. So far we have been able to fix the bugs, but if the plan doesn’t work, we would move back to the original firmware and build upon that.
  • Model Robustness: We reduced the data collection time of the wand to 1.5s (was 8s). This change was meant to make the device more user friendly. However, model performance became much worse than before even the validation accuracy remains high during training. We are working on capturing more data, and worse case we can increase the data collection time.
  • Verification and Validation: We are testing whether the 1.5s data collection window supports accurate gesture classification. Verification focuses on evaluating accuracy, and response time to ensure the model meets MVP requirement. For validation, we are running end-to-end tests to confirm that real user interactions with the wand reliably trigger correct IR outputs.

Design Changes & Impact

There is no design changes at this point.

Schedule Updates

No major changes were made and the schedule remains as is.

Progress / Photos

Wand (Everything put together):

Wand controlling the TV in 1300 coves with cloned signal

Please see the project GitHub.