1. Thompson sampling for networked control over unknown channels.
- Author
-
Liu, Wanchun, Leong, Alex S., and Quevedo, Daniel E.
- Subjects
- *
ACTUATORS , *SAMPLING methods - Abstract
We consider a networked control system where information is transmitted over lossy communication channels on both the uplink (sensor to controller) and downlink (controller to actuator). At each time instant, only one out of M different channels can be chosen for transmission. The statistics of the channels are unknown and must be learnt by observing transmission outcomes. We model this scenario as a multi-armed bandit type problem and investigate the use of Thompson sampling for carrying out this learning while simultaneously controlling the system. We derive necessary and sufficient conditions for stability of the controlled system. In addition, the notion of (accumulated) control regret is introduced and shown to scale logarithmically with time. Numerical comparisons show that Thompson sampling outperforms other channel selection methods such as periodic sampling and decaying ɛ -greedy. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF