Artur Majtczak and Maciek Matuszewski are looking to launch a Raspberry Pi Compute Module 4 (CM4) carrier board with a difference: it aims to offer broad expansion for motor control, a range of sensors, and voice control integrated with OpenAI's ChatGPT large language model (LLM).
"SaraKIT [is] the open source Raspberry Pi CM4 expansion board that enables advanced voice control and precise motor control," Majtczak writes of the device. "This versatile board boasts three sensitive microphones, enabling sound localization for voice recognition up to five meters [around 16 feet]. It also provides two independent BLDC [brushless DC] motor controllers, ensuring quiet, rapid, and precise motor control for various applications, including gimbal motors."
The board itself includes a range of integrated hardware, beyond the aforementioned triple microphone array and motor controllers: there are two inputs for rotary encoders, which can be repurposed as general-purpose input/output (GPIO) pins, and an STMicroelectronics LIS3DH 16-bit digital accelerometer.
The hardware bundle also features an STMicro LSM6DS3TR iNEMO six degrees of freedom (6DoF) inertial measurement unit (IMU) on a breakout board dubbed the SaraKIT Hub, included as standard, and a Microchip dsPIC33 16-bit microcontroller with 32kB of static RAM (SRAM). For further expansion, there are 11 GPIO pins carrying buses including UART, SPI, and I2C, with pulse-width modulation (PWM) support.
The SaraKIT aims to be a one-and-done design for a wide range of projects. For robotics, there's the IMU and motor controller — and the Hub board includes two MIPI Camera Serial Interface (CSI) ports for stereo vision applications.
For general-purpose electronics, there are the board's GPIO pins and the encoder inputs, plus the LSM6DS3TR can double as a temperature sensor. For audio work, the three microphones feed into a Microsemi ZL38063 and there are two six-watt amplifiers for output.
It's the audio side, though, which makes the SaraKIT stand out — as SaraAI hopes the see the board used to drive smart voice assistants powered by integration with OpenAI's ChatGPT large language model. "With SaraKIT," its creators claim, "you can build a ChatGPT-supported device housed in a 3D-printed casing with user face tracking and recognition capabilities.
"Just look at the camera before issuing a command," the company promises of its creation, "and your enhanced assistant will recognize your intent and know you’re addressing it, making interactions more natural and intuitive, like human conversation."
To prove its capabilities, the SaraKIT board has been used to power a range of demos including a smartphone-controlled LEGO car and a self-balancing LEGO robot, a gimbal-motor based "silent" pan-tilt stereo camera system, face and object detection and tracking, a sound locator, a depth-perception stereo vision system, and the ChatGPT-powered SaraEye.
The SaraKIT is now crowdfunding on Crowd Supply at $99 including the carrier board itself, the Hub board, and flex cable; buyers will have to supply their own Raspberry Pi Compute Module 4 board. Hardware is expected to ship in late February 2024, SaraAI says, with schematics published to GitHub under an unspecified open source license.