Spacecraft Docking Testbed
Neurocontrol technologies that can learn in near real-time the mass property of a spacecraft and its changes, and can compensate for thruster strength degradation, thruster failures, and uncontrolled venting (e.g. uncontrolled gas venting on Apollo 13 resulting from an onboard explosion), will make semi-autonomous / autonomous spacecraft navigation, rendezvous and docking safer, more accurate, faster, and more fuel efficient. Moreover, docking to a moving target will be possible.
The Smart Systems Research Lab at NASA Ames Research Center is developing adaptive neurocontrol technologies to safely, accurately and efficiently dock a spacecraft to a target under a wide range of difficult operating conditions. The MIR accident that occurred is an example of the kind of disaster that can take place if docking is not done optimally. Difficult operational scenarios include 1) docking a spacecraft when its thruster strengths are not well known, stuck or leaking, 2) docking a spacecraft whose mass property is not well known, 3) rendezvous and capturing a disabled spinning satellite, 4) docking a spacecraft to a spinning target, and (5) docking a spacecraft when some of its sensors become inoperable or fail. Operating a spacecraft under these conditions is a complex and dangerous task.
We are developing neurocontrol technologies to operate under these difficult operating conditions, and are applying them in various operational modalities, ranging from computer-aided joystick control, semi-automated docking, and fully automated docking.
Conventional automated docking approaches work well only when an accurate mathematical model of the spacecraft is available. In the operational scenarios described above, an accurate mathematical model is not available. The approach being applied is to use advanced identification technologies and adaptive neurocontrol to provide optimal control of the spacecraft. These development efforts merge adaptive neural network technologies with conventional feedback control techniques to handle problems in system identification and control of nonlinear systems. This approach does not require a mathematical model of the spacecraft a priori but instead it effectively learns an accurate model of the spacecraft from its behavior. The controller uses gathered data from a set of navigational sensors to quasi-statically learn an accurate model of the spacecraft performance. In this manner, the controller can control the spacecraft under a variety of changing conditions such as varying mass, changing center-of-mass location, thruster degradations/failures, uncontrolled venting, and sensor failures. Optimization methodologies are then used to achieve optimal performance.