Abstract
This paper compares the robustness of Infotaxis and Maximum A Posteriori (MAP) search strategies under parameter mismatch scenarios. Vergassola et al. (2007) originally introduced Infotaxis as a passive sensing strategy inspired by moth odor tracking. Infotaxis maximizes mutual information by choosing measurements to minimize the expected uncertainty in the target’s location. As a result, Infotaxis balances the exploration of its environment to gain new information with the exploitation of existing information. In contrast, MAP is a highly exploitative search strategy that makes decisions by relying on available information to guide the search. Recent studies extended Infotaxis’ application to active sensing, demonstrating its superiority over MAP in reducing search duration at detection probability (PD) below 0.8. To further investigate the robustness of Infotaxis and MAP, this study introduces parameter mismatches between the Bayesian update of the state vector and the sensor model. These mismatches occur when the estimated detection probability (ηD) differ from the sensor's actual PD. We evaluate three sensor calibration conditions — over-calibrated (PD > ηD), properly calibrated (PD = ηD), and under-calibrated(PD < ηD)—using 10 000 Monte Carlo simulations. Infotaxis consistently performs better than MAP under low PD (<0.85) scenarios. Infotaxis also excels in under-calibratedconditions. [Work supported by ONR MURI program.]