With a scrubbed test launch opportunity last year, the Skydart-I payload has seen a lot of firmware updates in the meantime. Over summer, a number of full stack simulations have highlighted the need for an upgraded navigation system for returning to a safe and reachable landing zone.
The previous system implementation followed simple logic, under the premise it was faster to implement, easier to test as well as be a sufficiently robust solution to the problem. More advanced simulations based on the latest numbers showed that while this would still be sufficient in the vast majority of cases, it was found that the worst case scenarios (a slow ascent, higher wind speeds in the upper atmosphere and a late decoupling from the ascent balloon) had the Skydart-I vehicle landing some distance short of the planned recovery point. A new implementation would be necessary to ensure the best possible chance of a successful mission.
At the heart of this new system is the Ordnance Survey ‘Elevation 50′ dataset. Once converted from ASCII format, the elevation data for the entire UK was then converted from OS National Grid reference co-ordinates to a WGS-84 (the standard used by most GPS receivers) compatible co-ordinate system the implementation could begin. The elevation data is then stitched together into a single binary data file, of which 48,000,000 sample points are selected and indexed into reference zones for use by the navigation system.
Most common navigation algorithms are designed to function on a 2D network, usually evaluating the time taken between different possible routes. Such implementations such as Djikstras’ and A* are common in computing today, usually powering satellite navigation devices and AI path-finding. For our purposes, however, such approaches would need to be modified to incorporate the expected altitude of the vehicle at a future point in time, as well as fully automating the formulation of each navigation run as no human interaction will be available on the day of the launch.
The problem is thus broken down into multiple stages, which are then parallelized across 2 of the 4 available CPU cores on the flight computer.
- Formulate simulation conditions based on current parameters, available resources and desired landing zone.
- Generate a simulation dataset at an appropriate resolution for the current situation.
- Generate a network of all possible routes from the current point to the destination.
- Compute the cost at each node of the network, taking into account the expected flight parameters at each point.
- Select the optimal solution from the computed dataset.
In practice, such a scheme works well, successfully avoiding all major obstacles in the simulations run thus far, with acceptable speed: a 1,500×1,500 node simulation space spanning a ~3,000km² area takes approximately 6 minutes to complete. Included below are the graphed datasets of one of the test scenarios run through the system showing the computed cost network and resulting path over the 3D dataset based on the ground features.
With this system now in place, Skydart-I continues to progress towards the upcoming launch later this year and we are now more confident than ever that it will be possible to recover an intact flight vehicle.