Active sensing theory: movement is for control, not just information

A new paper by Andrew Lamperski, Debojyoti Biswas, Eric Fortune, John Guckenheimer, Kathleen Hoffman, and Noah Cowan reframes active sensing as a control mechanism rather than a pure information-gathering process. Traditionally, active sensing has been understood as the expenditure of energy—often movement—to obtain sensory information. The team argues that this movement emerges inevitably from the combination of adaptive sensors, the coupling of movement and sensing, and the demands of task-level control. In other words, animals don't move their eyes or whiskers just to see or feel better; they do it to enable the control actions needed for survival.

The paper identifies two distinct behavioral modes observed across species: an 'explore' mode where animals produce dynamic movements to shape sensory feedback, and an 'exploit' mode where slower, compensatory movements directly serve task goals. While engineered systems with advanced sensors and actuators can outperform animals on specific metrics like force, speed, and precision, they still lack the graceful, robust behaviors common in biology. The authors suggest that incorporating a mode-switching control policy—one that toggles between exploration and exploitation—could bridge this gap, offering a path toward more adaptive and resilient robots.