G&C Networks - Deep architectures for real time optimal actions
Recent research on deep learning, a set of machine learning techniques able to learn deep architectures, has shown how robotic perception and action greatly benefits from these techniques. In terms of spacecraft navigation and control system, this suggests that deep architectures may be considered now to drive all or part of the on-board decision making system. In this project we prove that it is possible to train deep artificial neural networks to represent the optimal control action during different scenarios.
The resulting networks, called G&CNETS or gecnets, are able to safely perform the required task when trained accurately.
Our results allow for the design and validation of an on-board real time optimal control system able to cope with large sets of possible initial states while still producing an optimal response.
In the video above we show the logic of our research and we show, at the very end, an optimal landing of a Falcon 9 rocket completely driven by a G&CNET that takes its inputs from a second convolutional neural network trained to reconstruct the state (pose estimation) of the rocket.
The possiblity to have G&CNETS run on on embedded systems is currenty being studied using quadcopters platforms and in a joint research with TU Delft University and their MAV labs. Current investigations are looking into Neural Networks trained on time-optimal manoeuvres which are then run on a quadcopter at TU Delft. The Neural Network based controllers are being run in comparison with state-of-the-art controllers in order to evaluate their effectiveness. A publication will follow that summarises these results more precisely.
Sanchez-Sanchez, C., D. Izzo, and D. Hennes. 2016. “Learning the Optimal State-Feedback Using Deep Networks.” In 2016 IEEE Symposium Series on Computational Intelligence (SSCI). [link]
———. 2016. “Optimal Real-Time Landing Using Deep Networks.” In 6th International Conference on Astrodynamics Tools and Techniques (ICATT). [link]
Sanchez-Sanchez, C., and D. Izzo. 2018. “Real-Time Optimal Control via Deep Neural Networks: Study on Landing Problems.” Journal of Guidance Control and Dynamics 41 (5): 1122–35. [link]