characteristic
- With XL4005 switching power supply module (about 6 yuan), it can support 2S-3S lithium battery power supply
- 16-channel PWM servo control
- Acceleration sensor (MPU6050, optional patch welding or direct plug-in module)
- Use an additional dual-core ARM Cortex M0+ microcontroller RP2040 to run the TinyML machine learning program and support access to a microphone for speech recognition. Compared with the similarly priced STM32F103, it has richer memory resources and is more suitable for running machine learning. Edge Impulse has officially supported RP2040, and it is easy to find information on using RP2040 to run TensorFlow Lite online.
- All interfaces of the HI3861 module are introduced to facilitate the expansion of more functions in the future.
Physical verification has been carried out.
Code introduction
gait part
TROT gait forward and backward and left and right turns (series legs) have been realized. In the future, gyro stabilization function and WALK gait will be added.
RP2040 speech recognition part
The rp2040 firmware in the attachment has compiled code that can recognize the Chinese "handshake" keyword. Because the development time is limited, it is not possible to collect too much data (the training set is only more than five minutes) and perform tuning, and because most The training set was all recorded by myself, so the recognition effect on other people's voices is average (the test set recognition effect can be seen in the picture below). There is no practicality at this stage, mainly because of the model, but the performance of RP2040 is sufficient (16000Hz audio data per second requires about 400 milliseconds of inference and can be run in real time), and the trained model can accurately identify Key words.
Feature extraction
Neural network classification
The code and compiled firmware are in the attachment
grateful
The 3D printing file and quadruped gait algorithm come from Deng Ge’s MicroPython open source project, and part of the I2C driver code comes from Chuanzhi’s GenkiPi tutorial.
Physical picture