3D Robotics

More powerful Pixhawk 3 coming soon


The PX4/Dronecode team and Drotek have been working on the next generation of Pixhawk autopilots, and you can now see a preview of that with the Pixhawk 3 Pro. It's based on the new PX4 FMU4 Pro standard, which includes a full suite of next-generation sensors and and the more powerful STM32F469 processor. It's designed for the Dronecode/PX4 flight software, which is the current official Pixhawk standard.

The board is currently in developer release, but will be taken out of beta after testing is complete in the next month or two.

All details are here (and below):



FMUv4-PRO takes input from all of the Pixhawk stakeholders; end users, developers, researchers and manufacturing partners. Goals for this iteration of the platform are:

  • – An integrated, single-board flight controller for space constrained applications
  • – A modular multi-board flight controller for professional applications
  • – Sufficient I/O for most applications without expansion.
  • – Improved ease-of-use.
  • – Improved sensor performance
  • – Improved microcontroller resources (384 KB RAM, 2 MB flash).
  • – Increased reliability and reduced integration complexity.
  • – Reduced BoM and manufacturing costs.


Key design points

  • – All-in-one design with integrated FMU and optional IO and lots of I/O ports.
  • – Improved manufacturability, designed for simpler mounting and case design.
  • – Separate power supplies for FMU and IO (see power architecture section).
  • – Onboard battery backup for FMU and IO SRAM / RTC.
  • – Integration with two standard power bricks.


Technology upgrades

  • – Microcontroller upgrade to STM32F469; flash increases from 1MiB to 2MiB, RAM increases from 256KiB to 384KiB, more peripheral ports.
  • – ICM-20608, MPU9K integrated gyro / accelerometer / magnetometers.
  • – LIS3MDL compass (HMC5983 is now obsolete).
  • – Sensors connected via two SPI buses (one high rate and one low-noise bus)
  • – Two I2C buses
  • – Two CAN buses
  • – Voltage / battery readings from two power modules
  • – FrSky Inverter
  • – JST GH user-friendly connectors


I/O ports

  • – 6-14 PWM servo outputs (8 from IO, 6 from FMU).
  • – R/C inputs for CPPM, Spektrum / DSM and S.Bus.
  • – Analog / PWM RSSI input.
  • – S.Bus servo output.
  • – 6 general purpose serial ports, 2 with full flow control, 1 with separate 1A current limit, 1 with FrSky protocol inverter.
  • – Two I2C ports.
  • – Two external SPI ports (unbuffered, for short cables only).
  • – Two CANBus interfaces.
  • – Analog inputs for voltage / current of two batteries
  • – On-ground usage piezo buzzer driver.
  • – Sensor upgrade connector scheme
  • – High-power RGB LED.
  • – Safety switch / LED.


Mechanical Form Factor

  • – 71 x 49 x 23 mm (with case)
  • – 45g (with case)
  • – Standardized microUSB connector location
  • – Standardized RGB led location
  • – Standardized connector locations


System architecture

FMUv4-PRO continues the PX4FMU+PX4IO architecture from the previous generation, incorporating the two functional blocks in a single physical module.


PWM Outputs

Eight PWM outputs are connected to IO and can be controlled by IO directly via R/C input and onboard mixing even if FMU is not active (failsafe / manual mode). Multiple update rates can be supported on these outputs in three groups; one group of four and two groups of two. PWM signal rates up to 400Hz can be supported.

Six PWM outputs are connected to FMU and feature reduced update latency. These outputs cannot be controlled by IO in failsafe conditions. Multiple update rates can be supported on these outputs in two groups; one group of four and one group of two. PWM signal rates up to 400Hz can be supported.

All PWM outputs are ESD-protected, and they are designed to survive accidental mis-connection of servos without being damaged. The servo drivers are specified to drive a 50pF servo input load over 2m of 26AWG servo cable. PWM outputs can also be configured as individual GPIOs. Note that these are not high-power outputs – the PWM drivers are designed for driving servos and similar logic inputs only, not relays or LEDs.


Peripheral Ports

FMUv4-PRO recommends separate connectors for each of the peripheral ports (with a few exceptions). This avoids the issues many users reported connecting to the 15-pin multi-IO port on the original PX4FMU-PRO and allows single-purpose peripheral cables to be manufactured.

Five serial ports are provided. TELEM 1, 2 and 3 feature full flow control. TELEM4 can be switched into inverted mode by hardware and has no flow control. Serial ports are 3.3V CMOS logic level, 5V tolerant, buffered and ESDprotected.

The SPI ports are not buffered; they should only be used with short cable runs. Signals are 3.3V CMOS logic level, but 5V tolerant.

Two power modules (voltage and current for each module) can be sampled by the main processor.

The RSSI input supports either PWM or analog RSSI. CPPM, S.Bus and DSM/ Spektrum share now a single port and are auto-detected in software.

The CAN ports are standard CANBus; termination for one end of the bus is fixed onboard. .



The new ICM-20608 has been positioned by Invensense as higher-end successor of the MPU-6000 series. The software also supports the MPU-9250, which allows a very cost-effective 9D solution.

Data-ready signals from all sensors (except the MS5611, which does not have one) are routed to separate interrupt and timer capture pins on FMU. This will permit precise time-stamping of sensor data.

The two external SPI buses and six associated chip select lines allow to add additional sensors and SPI-interfaced payload as needed.

IMU is isolated from vibrations. 

Power Architecture

Key features of the FMUv4-PRO power architecture:

  • – Single, independent 5V supply for the flight controller and peripherals.
  • – Integration with two standard power bricks, including current and voltage sensing.
  • – Low power consumption and heat dissipation.
  • – Power distribution and monitoring for peripheral devices.
  • – Protection against common wiring faults; under/over-voltage protection, overcurrent protection, thermal protection.
  • – Brown-out resilience and detection.
  • – Backup power for IO in the case of FMU power supply failure.
  • – Split digital and analog power domains for FMU and sensors.


FMU and IO Power Supplies

Both FMU and IO operate at 3.3V, and each has its own private dual-channel regulator. In order to address issues seen with PX4v1 and noisy power supply connections, each regulator features a power-on reset output tied to the regulator’s internal power-up and drop-out sequencing.

The second channel of each dual regulator is switchable under software control. For FMU this is used to permit power-cycling the sensors (in case of sensor lockup), and for IO this will make it possible to perform the Spektrum binding sequence.


Power Sources

Power may be supplied to FMUv4-PRO via USB (no peripherals in this mode) or via the power brick ports. Each power source is protected against reverse-polarity connections and back-powering from other sources. Power spikes observed on the servo bus (up to 10V) led to the removal of the power-from-servo option, users wanting this feature can connect the servo rail with a cable containing a Zener diode to the 2nd power brick port.

The FMU + IO power budget is 250mA, including all LEDs and the Piezo buzzer. Peripheral power is limited to 2A total.


Power Brick Port

The brick port is the preferred power source for FMUv4-PRO, and brick power will be always be selected if it is available.


Servo Power

FMUv4-PRO supports both standard (5V) and high-voltage (up to 10V) servo power with some restrictions. IO will accept power from the servo connector up to 10V. This allows IO to fail-over to servo power in all cases if the main power supply is lost or interrupted. FMUv4-PRO and peripherals combined may draw up to 2A total.

Power is never supplied by FMUv4 to servos.


USB Power

Power from USB is supported for software update, testing and development purposes. USB power is supplied to the peripheral ports for testing purposes, however total current consumption must typically be limited to 500mA, including peripherals, to avoid overloading the host USB port.


Multiple Power Sources

When more than one power source is connected, power will be drawn from the highest-priority source with a valid input voltage.

In most cases, FMU should be powered via the power brick or a compatible offboard regulator via the brick port or servo power rail.

In desktop testing scenarios, taking power from USB avoids the need for a BEC or similar servo power source (though servos themselves will still need external power).



For each of the components listed, the input voltage ranges over which the device can be powered from each input is shown.

Brick ports Servo rail USB port
    FMU 4 – 5.7V no yes
    IO 4 – 5.7V 4 – 10V yes
    Peripherals 4 -5.7V, 2A max 4 – 5.7V, 250mA max yes, 500 mA max


Peripheral Power :

FMUv4-PRO provides power routing, over/under voltage detection and protection, filtering, switching, current-limiting and transient suppression for peripherals.

Power outputs to peripherals feature ESD and EMI filtering, and the power supply protection scheme ensures that no more than 5.5V is presented to peripheral devices. Power is disconnected from the peripherals when the available supply voltage falls below 4V, or rises above approximately 5.7V.

Peripheral power is split into two groups:

  • – TELEM 1 has a private 1A current limit, intended for powering a telemetry radio. This output is separately EMI filtered and draws directly from the USB / Brick inputs. Due to the noise induced by radios powering a radio from this port is not advised.
  • – All other peripherals share a 1A current limit and a single power switch. 

Each group is individually switched under software control.

The Spektrum / DSM R/C interface draws power from the same sources as IO, rather than from either of the groups above. This port is switched under software control so that Spektrum / DSM binding can be implemented. Spektrum receivers generally draw ~25mA, and this is factored into the IO power budget. S.Bus and CPPM receivers are supported on this rail as well.


Battery Backup :

Both the FMU and IO microcontrollers feature battery-backed realtime clocks and SRAM. The onboard backup battery has capacity sufficient for the intended use of the clock and SRAM, which is to provide storage to permit orderly recovery from unintended power loss or other causes of in-air restarts. The battery is recharged from the FMU 3.3V rail. 


Voltage, Current and Fault Sensing :

The battery voltage and current reported by the power brick can be measured by FMU. In addition, the 5V unregulated supply rail can be measured (to detect brown-out conditions). IO can measure the servo power rail voltage.

Over-current conditions on the peripheral power ports can be detected by the FMU. Hardware lock-out prevents damage due to persistent short-circuits on these ports. The lock-out can be reset by FMU software.

The under/over voltage supervisor for FMU provides an output that is used to hold FMU in reset during brown-out events.


EMI Filtering and Transient Protection :

EMI filtering is provided at key points in the system using high-insertion-loss passthrough filters. These filters are paired with TVS diodes at the peripheral connectors to suppress power transients.

Reverse polarity protection is provided at each of the power inputs.

USB signals are filtered and terminated with a combined termination/TVS array.

Most digital peripheral signals (all PWM outputs, serial ports, I2C port) are driven using feature series blocking resistors to reduce the risk of damage due to transients or accidental mis-connections.

Inputs / Outputs

InputOutput pixhawk pro by Drotek

E-mail me when people leave their comments –

You need to be a member of diydrones to add comments!

Join diydrones


  • Developer

    Yes, the whole point of EKF filters is to use multiple data sources to separate noise from actual data. As long as the signal to noise ratio is better then 50%, you will always get better results by adding more data points from the same time domain. After a while there are diminishing returns, but two or three is definitively better then one. Especially in this case as mentioned, where each sensor is selected to have different frequency characteristics.

  • Developer

    The real issue is PX4 can not process 3 IMU's at the same time due to it's infrastructure design where it processes IMU data in the IRQ thread. ArduPilot has more features and processes/fuses the IMU data in parallel in less CPU cycles and less RAM. Pixhawk 2.1 has 3 IMUs because ArduPilot *actually uses them* in a redundant and constructive way.

    Plus, the 3 IMUs on the Pixhawk 2.1 is there for a whole different reason: 1 is hard mounted and 2 vibrationally isolated giving you good flight data with a backup, plus vibration information about your frame that you can always log and review with the data spoiling normal flight data. This lacks that. Also, there are different brands of IMUs on the Pixhawk 2.1 with different characteristics which are sampled at different rates to help detect and remove vibration harmonics that may otherwise be missed by sampling IMU(s) at a single rate.

    When ArduPilot is finished removing it's PX4Firmware submodule it'll save a ton of flash space and reduce a ton of compile warnings and compile *way* faster.

    And @DROTEK I'm sorry to tell you that you're plain wrong. If a given EKF was able to compute 1000 IMUs it actually results in a better result than is using 1 IMU. EKFs eat data noise for breakfast. Adding more states/IMUs always helps as long as you have the resources to handle it.

  • It is good to have options. But the Pixhawk name is pretty meaningless now.
  • So glad to see this. PixHawk 2 is hard to get a hold of and not easy to setup compared to Pixhawk 1. Never liked the idea of the "cube" anyway.

  • No one is taking anybody for an idiot. It is a completely different way of tackling the problem by trying to take it at its root, i.e. MEMS temperature-dependant response. This method, as you say, has its own flaws, but what about trying it before claiming its uselessness? Let's try to be constructive.

    Drotek...ummm constructive you say after you call temperature control a hack in the first place!!!!

  • Developer

    Adding a third IMU makes sense for a lot of reasons.  You now have an odd man out voter (see space shuttle).  With only two you you have to cleverly "guess" which is the more reliable IMU at any given point in time.  Adding a third doesn't have to increase your noise there are numerous implementations to correctly address multiple attitude solutions from mems.

  • No one is taking anybody for an idiot. It is a completely different way of tackling the problem by trying to take it at its root, i.e. MEMS temperature-dependant response. This method, as you say, has its own flaws, but what about trying it before claiming its uselessness? Let's try to be constructive. The final idea is to have better product in the end. 

    This new MCU also has much more I/Os (great for integrating new sensors/payloads with other communication buses, this expands possibilities and is consistent with the open source/hardware logic) and its package is compatible with next ARM generations (like the F7 for instance). We'd be happy to know your expectations as to what you have in mind for these hardware evolutions. 

    Regarding the ICM20608, it is not as buggy and is actually pin to pin compatible with other devices like the ICM20602 that can be used as drop-in replacement.

  • Temperature control is a hack, a better way of doing things would be to log the response of different sensors in a large temperature range and adjust thermal compensation coefficients in accordance with that measurement.

    With all due respect, that is plain nonsense. I think you want to look up the thermal expansion responses of mechanical devices (which MEMS are ;)) upon temperature sweeps, and will discover that the actual response is history dependent, and very significantly so. Doing what you suggest is better than doing nothing (and needs to be repeated every once in a while), but to say that it is a "better way" is really close to incompetent.

    The "new" mcu is as performant as all the previous 32-bit ones. Since essentially 1.5 to 2 years all boards have 2Mb (rev 3). The "new" ICM-20608 is bugy and has been outdated by InvenSense after only ca. 2 years, and is explicitly not recommended for new designs (https://www.invensense.com/products/motion-tracking/6-axis/). So, the "more powerfull" is 50% more RAM, and more interfaces. It's kind of calling a car with 6 instead of 5 doors and a bit more space but the very same motor more powerfull than the 5 door variant.

    I think we all appreciate a lot your efforts and contributions, but I think we all think we are no fools :).

  • I think this is an eternal debate but what is the point in having three times the same IMU? What is the point in adding more noise to your model? That is not redundancy. It would make sense to have three different autopilots but not three times the same IMU because it does not really bring any increased value.

    Temperature control is a hack, a better way of doing things would be to log the response of different sensors in a large temperature range and adjust thermal compensation coefficients in accordance with that measurement. This effort has already been done on PX4 side (see there) and could easily be ported to Ardupilot. This method does not rely on an external physical component and ensures proper bias calculation/response over a large temperature range. In addition you don't have to wait for your IMU to reach the desired temperature. 

    The board comes with a new processor, with more RAM and more flash, important parameter when we know that current code has reached the 1MB previous limit. The idea behind this board was to take inputs from all users and make a real new design with improved I/Os, robustness and reliability.

  • Developer

    One less IMU that's not temperature controlled for more money?  How is this not a step backwards from the Pixhawk2?

This reply was deleted.