Controlling a Finless Rocket Through Neuroevolution

These two movies show JSBSim simulation of the Interorbital Systems, Inc. RSX-2 rocket with fins removed. In both movies, the amount of fuel is indicated by the red coloring of the rocket. The guard rail is 50' high. Alpha and Beta are the angles of the rocket wrt. its flight path. Flight time is shown on lower right.



1. Without active control
, i.e. with all engines at the same fixed power (613.25), the rocket is unstable: after only 2000', it gets out of alignment with its flight path, and there's no correcting force. As a result, the rocket explodes (although the simulation simply shows it tumbling wildly).
 

2. With neuroevolution control
, i.e. the evolved neural network controlling the power to the four engines, the rocket is stable: it flies straight until it runs out of fuel at the edge of the atmosphere; it then coasts and eventually reaches a 20 miles higher altitude than a similar rocket with fins.



Back to rocket control project page
Back to Neural Networks Research Group home page