Welcome to the Hardware repository of the Social Perceptive Lab for the MIDGE badge.

The Midge badge is a 55x35mm wearable PCB for in-the-wild social data collection. It features

• 2 microphones, with adjustable low and high cutoff frequencies
• 9 axis IMU
• SD card storage
• 300mAh battery
• BMD 300 processor with bluetooth low energy

Learn more about the project itself here.

Learn more about the firmware and software here.

Wearable badges are a cheap, convenient way to record high-fidelity signals from individuals in social interaction. We have the need for such a device in the SPC lab because we study the social dynamics of small crowds (~25-100 people) interacting during an event. Conference badges allow us to equip every participant to record individual audio, body movement signals and proximity information, all at a relatively low cost per device.

Cell phones can be useful in sensing crowds but differences in the sensors, fidelities and recording options / software available to different cell phone models are an issue for analysis. If cell phones are to be provided by the scientist, then cost becomes an issue in comparison with wearable badges.

Wrist wearables can be effective at capturing hand movements. While we could have designed our device with a wrist form factor, we think the badge form is more convenient to use and its position on the chest allows it to more reliably capture proximity and overall body movement through acceleration.

The open-source Rhythm Badge was designed with office environments in mind, where the goal was the collection of longitudinal data about a team over days or weeks. Proximity information and a low-frequency microphone allowed it to reconstruct the social network of team members. On the other hand our goal was to computationally analyse interactions both at the group and individual level from social signals (speech, movement, proximity) of subjects. This required higher fidelity sensors and a new device design.

• Enabled full audio recording, with a frequency up to 48KHz with an on-board switch to allow physical selection between high and low frequency.
• Added a 9-axis IMU to record pose. IMUs combine three tri-axial sensors: an accelerometer, a gyroscope, and a magnetometer. These measure acceleration, orientation, and angular rates respectively. The sensor information is combined on-chip by a Digital Motion Processor.
• Added an on-board SD card to directly store raw data, avoiding typical issues related to packet loss during wireless data transfer.

This project is unique because it improves upon the state-of-the-art wearable sensors measuring face-to-face human social interactions. The improvement points are catered towards more fine-grained and flexible capturing of data. For instance, the choice between high and low frequency audio capturing mode is provided at data acquisition and the experimenters can freely choose the frequency best suitable for them (for example, whether they have high-frequency audio for transcription or privacy-preserving low-frequency audio for detecting speech activities). Additionally, the inclusion of the on-board digital motion processor directly provides orientation estimation with the newly included full 9-DOF inertial motion unit (IMU). This allows future applications (e.g., real-time F-formation detection) to be built directly on top of the outputs of the DMP. The above design considerations and implementation exemplify the uniqueness and new advantages that Midges offer.

The design of the Midge is complete and 100 devices have been created. The device was successfully used for the ConfLab dataset. There it was used to record audio (1250Hz), IMU data (56Hz) and proximity data (1Hz) for 48 data subjects in the same space in a computer science conference. It has also been used to make smart drinking glasses and in several master and bachelor thesis.

The development of the device was contracted to Ioannis Protonotarios for the Socially Perceptive Computing Lab at Delft University of Technology. Ioannis took as base the latest release of the MIT Media Lab's Rhythm badge and implemented both hardware and software changes and updates. Testing of the device was done in conjunction with Socially Perceptive Computing Lab members. The whole process took 3 years and two iterations of the device's hardware. Additional firmware and software improvements have been developed by the SIPLab-EdgeAI at the Costa Rica Institute of Technology.

Absolutely. Some of the contributions that we current find most important in our application setting (sensing of small crowds) are: Improvements to audio sensing. Although our Midge was an improvement over the Rhythm badge in quality of audio sensing, audio quality is still an issue. Especially, the front-facing microphones of the device introduce significant cross-contamination. Potential solutions include exploring beam-forming, directional microphones or addition of headset support. Software improvements. There is significant room for improving the software of the device. The hub, which serves as monitoring and control centre for all connected Midges is especially relevant. The current command-line interface would benefit from being replaced by a user-friendly web-based interface that displays the status of connected midges and allows for quick bulk actions like eg. disabling sensors, starting/stopping recordings, etc.

Currently we are not aware of specific problems with the PCB, all functions work correctly (battery charging, cut-off switches, SD card, audio recording, IMU recording, status LEDs. If you do find a problem, please open an issue in the github repository.