Towards Oblivious Guidance Systems for Autonomous Vehicles

We consider encrypted guidance systems for straight-line path following. With cloud-computing technology, we can outsource computations in guidance systems to third-party providers, increasing scalability and enabling guidance-as-a-service. However, by remotely hosting a conventional guidance system on third-party infrastructure, we leak information such as the path of the vehicle. Potential customers may consider these leaks a serious breach of confidentiality, which limits the practical use of such systems. Therefore, to make cloud-based guidance systems viable, we would like to design guidance systems that we can host remotely without revealing confidential information to the host. To this end, we show that we can construct guidance systems that operate on encrypted position measurements, encrypted waypoints, and the bearing between each waypoint by using homomorphic encryption, effectively preventing the cloud host from identifying the vehicle's position. We show that the proposed guidance laws are locally exponentially stable and that the induced computational latency is appropriate for the real-time guidance of autonomous vehicles. Through field experiments, we demonstrate that an encrypted guidance system is practical and allows an unmanned surface vehicle to follow an encrypted path. The originality of this work lies in conceptualizing, designing, and experimentally validating an encrypted guidance system, unlike other studies that considered encrypted control systems.