Robotics and Neuroscience Applications

The assumption of prior knowledge of mean wind direction is a significant simplification, as real robots and animals must estimate wind direction from noisy local sensors. A clock-state controller with a one-dimensional internal state can achieve near-optimal plume tracking at a fraction of the computational cost of Bayesian POMDP methods, making it attractive for embedded systems. A concrete behavioral protocol for distinguishing clock-state from history-dependent navigation in insects is proposed: impose identical prolonged blanks after different prior odor histories and compare recovery trajectories. The two-Q architecture points toward a general design principle: a coarse local-density classifier combined with two complementary recovery programs captures most of the adaptability of a full history-dependent policy at a fraction of the memory cost.