MLF on tracking and control with Piksi


Drones may be used to great effect as ‘swarms’ in outdoor performances, creating beautiful aerial choreographies that make use of their unique aesthetic, at once robotic and organic. Such performances are currently achieved with only the highest-end drones and at huge cost – making this technology accessible only to companies with big marketing and production budgets. As there is extensive work being done on indoor drone-swarm control using motion capture rigs, we decided to focus research on outdoor, affordable, solutions which could potentially be used in hard to reach places, in order to extend the possibilities for this type of drone art work.

If you are a technologist, you might want to skip straight to the code behind the tracking and control research we did here.

Pre-programmed accurate flight for drones requires very accurate positional data that is generated at a high rate. It is essential both to know precisely where the drone is, and to train the machine learning algorithms onboard to fly as close as possible to the planned flight. For indoor flight, this is nearly always done with motion capture camera systems, as they are not complex to use and are very accurate. At the moment, no analogue option exists for outdoor flight.

GPS (Global Positioning System) is used in many off-the-shelf commercial drone systems to help stabilise outdoor flight. However GPS can only give an accuracy of roughly 3.5 meters. This would mean that for a choreographed drone show, you would need a distance of roughly 10 metres in between each drone, creating poor aesthetics.

So, we identified a potential solution in the Piksi RTK-GPS receiver from San Francisco based start-up Swift Navigation. The Piksi uses an existing technology known as real-time-kinetic differential GPS.  (Packages it in a roughly 50mm x 50mm x 5mm device [excluding antennas] requiring only 5 volts and 500mW of power.) Swift Navigation claims the accuracy of their Piksi system to be roughly 2cm, much more useful than the 3 metres offered by GPS.

Research Results

When the Piksi system had a stable connection, the results were as good as advertised, delivering an accuracy of roughly 2cm. This connection, however, was difficult to get, as it needs signal lock with more GPS satellites than traditional GPS does. But, two possible things could have caused this problem. First, Swift recommends their product be used only in areas with no surrounding tall obstructions (which could possibly interfere with the signal from GPS satellites), so whilst there were not many, the surrounding high-rise buildings may have interfered with signal needed. Second, one of the Piksi’s could have been faulty, leading to the connection instability. This could be solved with contact with Swift to try and solve these problems.

Additionally, the Piksi struggled to give positional data fast enough for very high precision flight. The framework used in the investigation of the Piksi (the open sourced firmware and messaging system written by the MSAV group at ETH Zurich) operates at 50Hz. Due to a handshake between the Piksi and the host computer that was written into Swift, the data rate couldn’t exceed 20Hz without losing connection. It is unclear if we could exceed 20HZ by removing the handshake, and if the frameowrk can operate at a high enough performance with a data rate of only 20Hz.

However, there were a few successes in the research. The Piksi software has been shown to work on multiple platforms reliably. The Swift Navigation libraries are all open source, written in both Python and C++ (computer programming languages), and are easy to use with existing software.

The positional data, when successfully determined and returned, was in an easy to use format. And, perhaps most importantly, all positional data for every individual Piksi in the tracking system is returned by each Piksi, meaning that no additional wireless communication is needed outside of what the Piksi requires. Furthermore, each Piksi does not need its own host computer.

In conclusion, no result in the research ruled out the Piksi from being what we hoped, but nothing confirmed it, either. Future research is needed, but only with a lot of input from Swift. Their insights into the issues we experienced, along with development of the device, could see the Piksi being a great outdoor tracking and control system.

Next page in toolkit: Case studies: ‘The Drone Aviary’