The rpc module on the OpenMV Cam allows you to connect your OpenMV Cam to another microcontroller or computer and execute remote python (or procedure) calls on your OpenMV Cam. The rpc module also allows for the reverse too if you want your OpenMV Cam to be able to execute remote procedure (or python) calls on another microcontroller or computer.
For computer control the rpc python module in this directory implements the OpenMV rpc protocol for control of an OpenMV Cam over USB VCP (i.e. a USB serial port) or over Ethernet/WiFi (i.e. over sockets).
The OpenMV Cam rpc library on the computer only depends on pyserial. All other modules used by it come installed with python. To get pyserial just do:
pip install pyserial
However, the examples depend on pygame so you need to install pygame too:
pip install pygame
Because the interface library is implemented in pure python with no external dependencies it works on Windows, Mac, and Linux.
pySerial provides support for pure USB virtual COM ports, USB to RS232/RS422/RS485/TTL COM ports, and standard RS232/RS422/RS485/TTL COM ports. Please use the rpc_usb_vcp_master and rpc_usb_vcp_slave for pure USB virtual COM port communication and rpc_uart_master and rpc_uart_slave for USB to RS232/RS422/RS485/TTL COM ports and standard RS232/RS422/RS485/TTL COM ports.
Pure hardware RS232/RS422/RS485/TTL COM ports should work using the rpc_uart_master and rpc_uart_slave interfaces without issues. However, FTDI like USB to serial converter chips may add unexpected latency to communication. In particular, FTDI chips have a latency timer which is used to buffer bytes for transmission over USB to improve bandwidth... but, that also increases the worse case latency of a single byte being sent over USB to 16 ms by default. You need to reduce the latency timer to 1 ms as detailed below or the interface library will fail to work:
- https://stackoverflow.com/questions/54230836/inconsistent-delay-in-reading-serial-data-in-python-3-7-2-using-pyserial
- https://projectgus.com/2011/10/notes-on-ftdi-latency-with-arduino/
You may use the RPC Interface Library over CAN using Kvaser hardware on Windows and Linux (Kvaser does not support Mac). You need to install the following:
- For Windows
- For Linux
Please checkout the following scripts for how to control your OpenMV Cam from the comptuer:
- Slow but Synchronus JPG Image Transfer
- Fast JPG Image Streaming
- Face Detection, April Tag Detection, Color Tracking, and more
You will need to edit the example code above to choose which interface you want to use (USB versus Ethernet/WiFi) and to play with the settings the scripts use. When you run the scripts make sure to save them first after editing them and then run them with python -u <script_name> to make sure that script output to stdio is not buffered.
In general, for the controller device to use the rpc library you will create an interface object using the rpc library. For example:
interface = rpc.rpc_usb_vcp_master("COM3")
This create a USB VCP interface to talk to your OpenMV Cam over COM3. For Mac and Linux you would pass some type of /dev/... device instead.
Once the interface is created you just need to do:
memory_view_object_result = interface.call("remote_function_or_method_name", bytes_object_argument)
And the rpc library will try to execute that "remote_function_or_method_name" on your OpenMV Cam. The remote function or method will receive the bytes_object_argument which can be up to 2^32-1 bytes in size. Once the remote method finishes executing it will return a memory_view_object_result which can also be up to 2^32-1 bytes in size. Because the argument and response are both generic byte containers you can pass anything through the rpc library and receive any type of response. A simple way to pass arguments is to use struct.pack() to create the argument and struct.unpack() to receieve the argument on the OpenMV Cam side. For the response, the OpenMV Cam may send a string object or json string as the result which the computer can then interpret. Most objects or lists returned from method calls on the OpenMV Cam generate valid json strings when you call str() on them.
As for errors, if you try to execute a non-existant function or method name on the OpenMV Cam the call method will return an empty bytes() object. If the rpc library failed to communicate with the OpenMV Cam the rpc library will return None.
To keep things simple the rpc library doesn't maintain a connection between the master and slave devices. The call method encapsulates trying to connect to the OpenMV Cam, starting execution of the remote function or method, and getting the result.
Now, on the OpenMV Cam side of things you have to create an rpc interface to communicate with the computer. This looks like:
interface = rpc.rpc_usb_vcp_slave()
This will create the interface layer on the OpenMV Cam (this needs to be done on a script running on the OpenMV Cam - see the Remote Control example scripts).
Once you create the slave interface you then need to register call backs that the master can call with the interface object.
def remote_function_or_method_name(memoryview_object_argument):
<lots of code>
return bytes_object_result
interface.register_callback(remote_function_or_method_name)
You may register as many callbacks as you like on the OpenMV Cam that the computer can call. Finally, once you are done registering callbacks you just need to execute:
interface.loop()
On the OpenMV Cam to start the rpc library up and begin listening for the computer. Note that the loop() method does not return. Also, to make your OpenMV Cam more robust against errors you may want to wrap the loop() with try: and except: for whatever exceptions might be thrown by your callback methods. The rpc library will not generate any exceptions itself. Note: passing large data structures around (like jpeg images) can potentially exhaust the heap on the OpenMV Cam and generate MemoryError exceptions.
And that is it! The rpc library is designed to be simple to use. It was designed to allow remote control of the OpenMV Cam by a computer or microcontroller so there are also interfaces for control of your OpenMV Cam over CAN, I2C, SPI, and UART.
Please see the example scripts above for starting code on how to use the rpc library. The below API documents the public interface of the library.
The rpc base class is reimplemented by the rpc_master and rpc_slave classes to create the master and slave interfaces. It is a pure virtual class and not meant to be used directly.
This method is meant to be reimplemented by specific interface classes of rpc_master and rpc_slave. It should fill the buff argument which is either a bytearray or memoryview object of bytes from the interface equal to the length of the buff object in timeout_ms milliseconds. On timeout this method should return None. Note that for master and slave synchronization this method should try to always complete in at least timeout_ms milliseconds and not faster as the rpc_master and rpc_slave objects will automatically increase the timeout_ms to synchronize.
This method is meant to be reimplemented by specific interface classes of rpc_master and rpc_slave. It should send data bytes on the interface within timeout_ms milliseconds. If it completes faster than the timeout that is okay. No return value is expected.
This method is meant to be called directly. After synchronization of the master and slave on return of a callback stream_reader may be called to receive data as fast as possible from the master or slave device. call_back will be called repeatedly with a bytes_or_memory_view argument that was sent by the stream_writer. call_back is not expected to return anything. queue_depth defines how many frames of data the stream_writer may generate before slowing down and waiting on the stream_reader. Higher queue_depth values lead to higher performance (up to a point) but require the stream_reader to be able to handle outstanding packets in its interface layer. Note that computers typically do not buffer much more than 4KB of data in device driver buffers. If you make the queue_depth larger than 1 then call_back should return very quickly and not block. Otherwise, you should implement a multi-thread architecture to process the received data so that stream_reader is always executing and moving data out of device driver buffers into larger user memory buffers. Finally, read_timeout_ms defines how many milliseconds to wait to receive the bytes_or_memory_view payload per call_back.
On any errors stream_reader will return. The master and slave devices can try to setup the stream again afterwards to continue as demonstrated in Fast JPG Image Streaming.
If you need to cancel the stream_reader just raise an exception in the call_back and catch it. The remote side will automatically timeout.
This method is meant to be called directly. After synchronization of the master and slave on return of a callback stream_writer may be called to send data as fast as possible from the master or slave device. call_back will be called repeatedly and should return a bytes_or_memory_view object that will be sent to the stream_reader. call_back should not take any arguments. Finally, write_timeout_ms defines how many milliseconds to wait to send the bytes_or_memory_view object returned by call_back.
On any errors stream_writer will return. The master and slave devices can try to setup the stream again afterwards to continue as demonstrated in Fast JPG Image Streaming.
If you need to cancel the stream_writer just raise an exception in the call_back and catch it. The remote side will automatically timeout.
The rpc_master is a pure virtual class and not meant to be used directly. Specific interface classes should reimplement rpc_master.
Executes a remote call on the slave device. name is a string name of the remote function or method to execute. data is the bytes like object that will be sent as the argument of the remote function or method to exeucte. send_timeout defines how many milliseconds to wait while trying to connect to the slave and get it to execute the remote function or method. Once the master starts sending the argument to the slave deivce send_timeout does not apply. The library will allow the argument to take up to 5 seconds to be sent. recv_timeout defines how many milliseconds to wait after the slave started executing the remote method to receive the repsonse. Note that once the master starts receiving the repsonse recv_timeout does not apply. The library will allow the response to take up to 5 seconds to be received.
Note that a new packet that includes a copy of data will be created internally inside the rpc library. You may encounter memory issues on the OpenMV Cam if you try to pass very large data arguments.
The rpc_slave is a pure virtual class and not meant to be used directly. Specific interface classes should reimplement rpc_slave.
Registers a call back that can be executed by the master device. The call back should take one argument which will be a memoryview object and it should return a bytes() like object as the result. The call back should return in less than 1 second if possible.
After you execute loop() it is not possible to execute long running operations outside of the rpc library. schedule_callback allows you to break out of the rpc library temporarily after completion of an call back. You should execute schedule_callback during the execution of an rpc call back method to register a new non-rpc call back that will be executed immediately after the successful completion of that call back you executed schedule_callback in. The function or method should not take any arguments. After the the call back that was registered returns it must be registered again in the next parent call back. On any error of the parent call back the registered call back will not be called and must be registered again. Here's how to use this:
def some_function_or_method_that_takes_a_long_time_to_execute():
<do stuff>
def normal_rpc_call_back(data):
<process data>
interface.schedule_callback(some_function_or_method_that_takes_a_long_time_to_execute)
return bytes(response)
interface.register_callback(normal_rpc_call_back)
interface.loop()
schedule_callback in particular allows you to use the get_bytes and put_bytes methods for cut-through data transfer between one device and another without the cost of packetization which limits the size of the data moved inside the rpc library without running out of memory on the OpenMV Cam.
The loop call back is called every loop iteration of loop(). Unlike the schedule_callback() call back this call back stays registered after being registered once. You can use the loop call back to blink an activity LED or something like that. You should not use the loop call back to execute any blocking code as this will get in the way of polling for communication from the master. Additionally, the loop call back will be called at a variable rate depending on when and what call backs the master is trying to execute. Given this, the loop call back is not suitable for any method that needs to be executed at a fixed frequency.
On the OpenMV Cam, if you need to execute something at a fixed frequency, you should setup a timer before executing loop() and use a timer interrupt based callback to execute some function or method at a fixed frequency. Please see how to Write Interrupt Handlers for more information. Note: The mutex library is installed on your OpenMV Cam along with the rpc library.
Starts execution of the rpc library on the slave to receive data. This method does not return (except via an exception from a call back). You should register all call backs first before executing this method. However, it is possible to register new call backs inside of a call back previously being registered that is executing.
recv_timeout defines how long to wait to receive a command from the master device before trying again. send_timeout defines how long the slave will wait for the master to receive the call back response before going back to trying to receive. The loop call back will be executed before trying to receive again.
Creates a master implementation of the rpc library to communicate over a hardware RS232/RS422/RS485 or TTL COM port. port is the string name of the serial port. baudrate is the bits per second to run at.
Please use the rpc_usb_vcp_master to talk to the OpenMV Cam over its USB interface. This class is for talking to the OpenMV Cam over its hardware UART.
If you are using this class with USB to serial converters you will likely experience intermittent synchronization issues. The RPC library uart interface on the OpenMV Cam is designed to work with real-time UARTs like hardware RS232/RS422/RS485 COM ports on PCs and TTL UARTS on single board computers like the RaspberryPi and Beaglebone. That said, the interface library can handle this and will continue to work.
Creates a slave implementation of the rpc library to communicate over a hardware RS232/RS422/RS485 or TTL COM port. port is the string name of the serial port. baudrate is the bits per second to run at.
Please use the rpc_usb_vcp_slave to talk to the OpenMV Cam over its USB interface. This class is for talking to the OpenMV Cam over its hardware UART.
If you are using this class with USB to serial converters you will likely experience intermittent synchronization issues. The RPC library uart interface on the OpenMV Cam is designed to work with real-time UARTs like hardware RS232/RS422/RS485 COM ports on PCs and TTL UARTS on single board computers like the RaspberryPi and Beaglebone. That said, the interface library can handle this and will continue to work.
Creates a master implementation of the rpc library to communicate over a USB VCP (virtual COM port). port is the string name of the serial port.
Communication over USB is the most reliable and high speed way to connect an OpenMV Cam to a computer. However, as the OpenMV Cam has only one VCP port we recommend that you fully debug your script using the rpc_usb_vcp_slave on your OpenMV using another rpc interface if possible as you will not be able to get error messages off the OpenMV Cam easily. For example, connecting the rpc_usb_vcp_master to the rpc_uart_slave on your OpenMV Cam at 115200 BPS over a USB-to-Serial adapter when debugging will make your life far easier. Alternatively, use the Ethernet or WiFi interface when debugging.
Creates a slave implementation of the rpc library to communicate over a USB VCP (virtual COM port). port is the string name of the serial port.
Communication over USB is the most reliable and high speed way to connect an OpenMV Cam to a computer. However, as the OpenMV Cam has only one VCP port we recommend that you fully debug your script using the rpc_usb_vcp_master on your OpenMV using another rpc interface if possible as you will not be able to get error messages off the OpenMV Cam easily. For example, connecting the rpc_usb_vcp_slave to the rpc_uart_master on your OpenMV Cam at 115200 BPS over a USB-to-Serial adapter when debugging will make your life far easier. Alternatively, use the Ethernet or WiFi interface when debugging.
Creates a master implementation of the rpc library to communicate over WiFi or Ethernet. slave_ip is the IPV4 address of the rpc slave device. my_ip can be "" which binds the master to any interface adapter to communicate to the slave. If my_ip is not "" then it should be an IP address on the same subnet as slave_ip. port is a free port to use for UDP and TCP traffic.
Creates a slave implementation of the rpc library to communicate over WiFi or Ethernet. my_ip can be "" which binds the slave to any interface adapter to communicate to the master. If my_ip is not "" then it should be an IP address on the same subnet as the master. port is a free port to use for UDP and TCP traffic.
Creates a master implementation of the rpc library to communicate over CAN using Kvarser hardware. channel is the Kvarser channel to use. message_id is the 11-bit CAN ID to use for data transport. bit_rate is the CAN bus rate. sampling_point is the percentage sample point on the CAN bus.
Creates a slave implementation of the rpc library to communicate over CAN using Kvarser hardware. channel is the Kvarser channel to use. message_id is the 11-bit CAN ID to use for data transport. bit_rate is the CAN bus rate. sampling_point is the percentage sample point on the CAN bus.