Towards the Embodiment of Active Sensing in Robotic Systems

Active sensing enables robots to interact dynamically with their environment by optimizing sensory information flow. However, replicating this in robotic systems poses challenges, particularly in developing metrics that account for sensor limitations and actuation noise, as well as integrating active sensing into control architectures. This paper presents an overview of active sensing integration, explaining a metric based on the Constructibility and Reachability Gramians and showing a methodology for incorporating active sensing into task-oriented control using a modified Lyapunov-based Model Predictive Control framework. The approach is validated through simulations on a unicycle vehicle.