Autonomous rendezvous system overcomes simulated spacecraft failures in lab tests

There is a problem in low Earth orbit, and it's growing. According to the European Space Agency, there are about 19,440 satellites on orbit, 3,540 of which are nonfunctioning and/or not emitting a signal to warn other spacecraft of their location. The need to service these sleepwalking satellites in space will only increase as more are launched.

"There are also over 8,000 metric tons (8,800 tons) of debris, none of which is cooperative, in the sense that they cannot thrust or send a signal," said Minduli Wijayatunga, a new aerospace engineering faculty member in The Grainger College of Engineering, University of Illinois Urbana-Champaign.

Although most of Wijayatunga's work can be applied to both debris and spacecraft, a recent study targets a spacecraft that needs repair but has lost the ability to communicate. The servicer must approach it with no help from the target, which is like an unconscious patient.

Wijayatunga said the repairs need to be done autonomously because of the time delay in communications between a repair spacecraft and Earth and the limited windows of time when communication is even possible.

The study is published in the journal Acta Astronautica.

Planning rendezvous around sunlight

"These spacecraft must be able to track the trajectories of uncooperative objects autonomously but also make go/no-go decisions without waiting for the ground to decide," she said. "Suppose your thruster fails or you encounter high-order orbital perturbations that push you away from your nominal trajectory. In those instances, how do you make those decisions?"

In response to this question, Wijayatunga created a framework for close-range rendezvous trajectory design and guidance and demonstrated it in a lab. It's called CORTEX, which stands for Convex Optimization for Rendezvous Trajectory Execution.

"CORTEX takes into account eclipses, timing the rendezvous when there is an optimal window of sunlight to see the target, complete your close-proximity operations and depart," she said. "For my Ph.D., I did a project on trajectory design and guidance using convex optimization for the transfer phases. Eclipses matter there too; they can limit how much you can thrust. But they don't compromise safety to the same extent, so you don't need to replan or hold position because of them.

"I wanted to bring that same idea into developing CORTEX and expand it, taking into consideration sunlight windows so we don't have to operate in darkness. That, and the replanning versus abort architecture, are novel in CORTEX."

Recovering from failed thrust and drift

To stress-test the hardware experiments in the lab, the servicer spacecraft was manually pushed to represent a perturbation that sends it off course from its target, thrusters were turned off during periods of time they were supposed to be on, and state and thrust errors were artificially increased to simulate unexpected situations like solar storms.

She explained that CORTEX decides whether to replan a new trajectory and approach the target or abort the approach, retreat to a passively safe orbit and retry the approach.

"CORTEX has trajectory design and guidance, so if you deviate significantly from your nominal mission trajectory, it replans the trajectory and reapproaches unless you are in severe violation of your keep-out zone, in which case it returns to a passively safe orbit. It's an algorithm to recover from the unexpected. We tried instances where you're supposed to use your thrusters, but you can't for whatever reason. And CORTEX was very robust to that."

"In some of the hardware tests, if the servicer violates the approach corridor, it leaves the keep-out zone and retreats to a safe position, and in others, the servicer is able to recover and carry on with its mission," she said.

Missing the target and regrouping for another approach is also an autonomous decision.

Extending autonomy to earlier rendezvous

Wijayatunga said she has a number of different projects that have stemmed from this work.

"One of the final projects in my class is using reinforcement learning to replan the phase that happens right before the close rendezvous. Far rendezvous happens when you see your target at a range of from 30 kilometers (19 miles) down to 3 kilometers (1.9 miles)."