Active Sensing

Active sensing studies how robots and other agents can select observations or sensing actions instead of relying on fixed, continuous measurements. Common approaches use Bayesian information gain, uncertainty-guided view selection, partially observable decision processes, and reinforcement learning to plan camera, tactile, acoustic, or other sensor actions. Applications range from robotic exploration and manipulation to medical imaging, environmental monitoring, and scientific measurement, often under limits on energy, time, or interaction. A central challenge is that reducing uncertainty does not always improve task success: methods must account for feasibility, latency, model error, and risk, and evaluate both evidence gained and downstream outcomes.