GAVONTS - The 'Lander Example'

As an example to illustrate the power of GAVONTS, we have used it to create a spacecraft lander control system. To achieve this we have written a descent and landing simulator that accepts engine commands and tracks the resulting path of the spacecraft down to a planetary surface.

The objective of the control system is to land the spacecraft at a chosen target point on the surface, with as low an impact velocity as possible, and having used as little fuel as possible; and to do this from a range of initial de-orbit positions, altitudes and approach velocities.

The descent simulator provides the control system with a basic set of information, including the current position and velocity of the spacecraft in 3-D space, and the remaining amount of fuel. Based on this information, the control system algorithm then returns engine firing commands to the simulator, which then computes and adjusts the descent trajectory accordingly.

The GAVONTS framework manages the descent simulator, together with an evolving population of control system models. As the process of simulated evolution progresses, the population becomes progressively 'fitter'. New and improved control system models, and better descent strategies, evolve spontaneously.

GAVONTS solves this problem relatively easily. In this example, after only 5000 generations of simulated evolution, a control system had evolved that could successfully land the spacecraft from a wide range of initial configurations, and with almost pinpoint accuracy.

To illustrate this, we have created a small virtual reality simulation movie clip that shows the progress of a lander descending under the direction of a control system algorithm generated by GAVONTS. To play this movie clip, click on the button at the left of this page - which launches a 'popup' containing a QuickTime movie clip.

A Key Point

It is important to emphasise that we have not guided GAVONTS in solving this problem; the solution has been created and evolved spontaneously using virtual natural selection. GAVONTS has produced this solution on its own, without any human guidance.

Notes on the Descent Simulation Movie Clip

In order to reduce the size of this movie clip, we have speeded it up by 8 times, and also used a relatively low 'frame rate'. This obviously increases the apparent rate of descent, and degrades the quality of the result. However the 'real time' version, with video quality frame rate, is >100 MB in size.

The movie clip shows the lander progressing out of orbit from an altitude of 150km, with an initial descent rate of 500mS^-1. The surface geometry for the planet shown in the simulation was created using an algorithm designed by us, and is intended only for the purposes of illustration, rather than to be of cgi quality.

The control system algorithm that was used here was obtained from GAVONTS after 5000 generations. This solution can consistently land the spacecraft, from a wide range of different starting configurations, within a few metres of the desired target, and with impact velocities of only a few mS^-1.

After 10,000 generations, even better performance is achieved; in particular: less fuel is used, and the engine burn profiles are much smoother.