In this post, I will explain the deployment of edge impulse to Arduino.

“You can deploy your impulse to any device. This makes the model run without an internet connection, minimizes latency, and runs with minimal power consumption.”

Please check this website: https://docs.edgeimpulse.com/docs/running-your-impulse-arduino

  • Go to your project from Edge Impulse, click Deployment.
  • Select Arduino from Deploy your impulse part.
  • Select Arduino Nano 33 BLE Sense from Build Firmware.
  • Select Quantized Version and click analyze.
  • Build your model.

It gives me arduino_motion-nano-33-ble-sense-v1.zip which includes flash.bat file. Reboot arduino and run flash.bat. Lunch command prompt run $edge-impulse-run –> This starts live classification.

But, this is not what I want. I would like to download impulse library.

I click build again without selecting Arduino Nano 33 BLE Sense from Build Firmware, 
it gives me ei-arduino-motion.zip this time.
  • Open Arduino IDE.
  • Sketch –> include licbrary –> Add zip –> select arduino-motion.zip
  • Tools –> board –> Arduino Mbed OS Nano Boards –> select Arduino Nano 33 BLE
  • File –> examples –> Arduino Motion Inferencing –> select Arduino Nano 33 BLE Sense Accelerometer
  • Tools –> Port –> Select Arduino COM
  • Click Upload

I encountered to this error:

Arduino_LSM9DS1.h: No such file or directory

Please check this website, I found it really heplpful: https://www.programmingelectronics.com/no-such-file-error/

I installed Arduino_LSM9DS1 from:

Open Arduino IDE: sketch --> include library --> manage libraries --> Arduino_LSM9DS1

Upload completed with the errors below:

`C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\ConvolutionFunctions\arm_convolve_HWC_q7_RGB.c: In function 'arm_convolve_HWC_q7_RGB':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\ConvolutionFunctions\arm_convolve_HWC_q7_RGB.c:125:25: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
                         *__SIMD32(pBuffer) = 0x0;
                         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\ConvolutionFunctions\arm_convolve_HWC_q7_RGB.c:158:25: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
                         *__SIMD32(pBuffer) = __PKHBT(bottom.word, top.word, 0);
                         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_nn_mult_q15.c: In function 'arm_nn_mult_q15':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_nn_mult_q15.c:98:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pDst)++ = __PKHBT(out2, out1, 16);
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_nn_mult_q15.c:99:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pDst)++ = __PKHBT(out4, out3, 16);
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_nn_mult_q7.c: In function 'arm_nn_mult_q7':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_nn_mult_q7.c:80:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pDst)++ = __PACKq7(out1, out2, out3, out4);
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_q7_to_q15_reordered_no_shift.c: In function 'arm_q7_to_q15_reordered_no_shift':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_q7_to_q15_reordered_no_shift.c:106:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pDst)++ = in2;
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\NNSupportFunctions\arm_q7_to_q15_reordered_no_shift.c:107:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pDst)++ = in1;
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\PoolingFunctions\arm_pool_q7_HWC.c: In function 'compare_and_replace_if_larger_q7':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\PoolingFunctions\arm_pool_q7_HWC.c:78:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pIn)++ = in.word;
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\PoolingFunctions\arm_pool_q7_HWC.c: In function 'accumulate_q7_to_q15':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\PoolingFunctions\arm_pool_q7_HWC.c:122:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pCnt)++ = __QADD16(vo1, in);
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\PoolingFunctions\arm_pool_q7_HWC.c:125:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pCnt)++ = __QADD16(vo2, in);
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\dsp\image\processing.cpp: In function 'int ei::image::processing::yuv422_to_rgb888(unsigned char*, const unsigned char*, unsigned int, ei::image::processing::YUV_OPTIONS)':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\dsp\image\processing.cpp:73:32: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
     for (unsigned int i = 0; i < in_size_pixels; ++i) {
                              ~~^~~~~~~~~~~~~~~~
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\tensorflow\lite\core\api\op_resolver.cpp: In function 'TfLiteStatus tflite::GetRegistrationFromOpCode(const tflite::OperatorCode*, const tflite::OpResolver&, tflite::ErrorReporter*, const TfLiteRegistration**)':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\tensorflow\lite\core\api\op_resolver.cpp:34:20: warning: comparison is always false due to limited range of data type [-Wtype-limits]
       builtin_code < BuiltinOperator_MIN) {
       ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~

Sketch uses 153392 bytes (15%) of program storage space. Maximum is 983040 bytes.
Global variables use 48248 bytes (18%) of dynamic memory, leaving 213896 bytes for local variables. Maximum is 262144 bytes.
Device       : nRF52840-QIAA
Version      : Arduino Bootloader (SAM-BA extended) 2.0 [Arduino:IKXYZ]
Address      : 0x0
Pages        : 256
Page Size    : 4096 bytes
Total Size   : 1024KB
Planes       : 1
Lock Regions : 0
Locked       : none
Security     : false
Erase flash

Done in 0.000 seconds
Write 153400 bytes to flash (38 pages)
[==============================] 100% (38/38 pages)
Done in 6.438 seconds` `

Open Serial Monitor for Live Classification. Despite the errors, live classification is working well:)

Starting inferencing in 2 seconds...
Sampling...
Predictions (DSP: 21 ms., Classification: 0 ms., Anomaly: 0 ms.): 
    circle: 0.00000
    idle: 0.03516
    left-right: 0.00000
    up-down: 0.96484

Starting inferencing in 2 seconds...
Sampling...
Predictions (DSP: 20 ms., Classification: 0 ms., Anomaly: 0 ms.): 
    circle: 0.91406
    idle: 0.00000
    left-right: 0.08594
    up-down: 0.00000

Starting inferencing in 2 seconds...
Sampling...
Predictions (DSP: 20 ms., Classification: 0 ms., Anomaly: 0 ms.): 
    circle: 0.00000
    idle: 0.99219
    left-right: 0.00781
    up-down: 0.00000

Let’s open the metadata:

C:\Users\pelin\Documents\Arduino\libraries -->  arduino_motion_inferencing --> src --> model_parameters

Open model_metadata.h.

Here is the error:

#error "Cannot use full TensorFlow Lite with EON"

I installed Arduino TensorflowLite via:

Open Arduino IDE: sketch --> include library --> manage libraries --> Arduino TensorflowLite.

Start upload again. Same error :(

C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\CMSIS\NN\Source\PoolingFunctions\arm_pool_q7_HWC.c:125:9: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
         *__SIMD32(pCnt)++ = __QADD16(vo2, in);
         ^
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\dsp\image\processing.cpp: In function 'int ei::image::processing::yuv422_to_rgb888(unsigned char*, const unsigned char*, unsigned int, ei::image::processing::YUV_OPTIONS)':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\dsp\image\processing.cpp:73:32: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
     for (unsigned int i = 0; i < in_size_pixels; ++i) {
                              ~~^~~~~~~~~~~~~~~~
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\tensorflow\lite\core\api\op_resolver.cpp: In function 'TfLiteStatus tflite::GetRegistrationFromOpCode(const tflite::OperatorCode*, const tflite::OpResolver&, tflite::ErrorReporter*, const TfLiteRegistration**)':
C:\Users\pelin\Documents\Arduino\libraries\arduino_motion_inferencing\src\edge-impulse-sdk\tensorflow\lite\core\api\op_resolver.cpp:34:20: warning: comparison is always false due to limited range of data type [-Wtype-limits]
       builtin_code < BuiltinOperator_MIN) {
       ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~

Yet, when I open:

File -> examples -> arduino_motion_inferencing -> Arduino Nano 33 BLE Sense Accelerometer Contionuous

Upload and open serial monitor:

Predictions (DSP: 28 ms., Classification: 0 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 1 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 0 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 0 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 1 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 1 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 1 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 1 ms., Anomaly: 0 ms.): idle  [ 0, 10, 0, 0, 0, 0, ]
Predictions (DSP: 28 ms., Classification: 0 ms., Anomaly: 0 ms.): uncertain  [ 3, 0, 4, 0, 3, 0, ]
Predictions (DSP: 26 ms., Classification: 0 ms., Anomaly: 0 ms.): uncertain  [ 3, 0, 4, 0, 3, 0, ]
Predictions (DSP: 26 ms., Classification: 1 ms., Anomaly: 0 ms.): uncertain  [ 4, 0, 4, 0, 2, 0, ]
Predictions (DSP: 28 ms., Classification: 0 ms., Anomaly: 0 ms.): uncertain  [ 5, 0, 4, 0, 1, 0, ]
Predictions (DSP: 28 ms., Classification: 1 ms., Anomaly: 0 ms.): uncertain  [ 6, 0, 3, 0, 1, 0, ]
....
Predictions (DSP: 28 ms., Classification: 1 ms., Anomaly: 0 ms.): up-down  [1, 0, 1, 8, 0, 0, ]

No errors!

First 4 numbers in the list represents:

  • 1: circle
  • 2: idle
  • 3: left_right
  • 4: up_down

Key Words

Tiny ML, Arduino, Edge Impulse, Deployment