Spacecraft autonomy has always played an important role in the exploration of the solar system and beyond and our reliance on increasingly independent spacecraft will only grow. There will be varying levels of autonomy depending on the task at hand. Immediate applications of increased autonomy could possibly be onboard anomaly detection, health and resource management. Followed by more complex decision making and guidance, navigation & controls. 

I foresee the near-future of spacecraft autonomy being more model-driven approaches, as it's historically been, as opposed to data-driven models like deep learning. There are two reasonings for that:
 1. Deep learning is very compute intensive, and often spacecrafts are highly computationally resource constrained. 
 2. Deep learning lacks explainability. A neural network is considered a black box, and if we want verification & validation of the algorithms, we need to be able to see into this black box.

As access to space continues to be democratized, I see a possibility where CubeSats could actually be the propeller of more novel data-driven autonomy algorithms. The most recent [MarCo][1] CubeSat missions to Mars could entice the industry to send more of these spacecrafts. And as a result, take more risks in the type of autonomy is deployed. The other side to this coin is that physics-based simulation engines continue to get better and better. This will allow researchers to apply creative autonomy techniques to highly-realistic environments. Not only that, there is a lot of research going on in the area of synthetic data. This could be highly beneficial to generate high quality data that a rover's deep neural net could be trained on prior to being deployed. 
Check out [this paper][2] on using deep reinforcement learning for autonomous imaging and mapping of small bodies. 

Also [this paper][3] on using reinforcement learning for motion planning of hopping rovers.

[Check out this great survey on AI trends in GN&C][4]

Sorry if this is kind of a rambling response. There are a lot of great opportunities for spacecraft autonomy and a lot of factors that come into play when deploying increasingly complex algorithms. Ultimately, the future of spacecraft autonomy looks very bright.