Optimization and learning applied to control and modeling of mechatronic systems.

Abstract

The source seeking is a relevant topic on autonomous robotics. In a few words, it consists of seeking a scalar signal source position with only local information on base X-space. In such, the seek agent, for instance, a mobile tracking A-robot, samples by hypothesis C ∞ -class source signal ϕ-map constrained by hull (A) ⊆ X. Among available seeking methods, this work utilizes the barycenter method, first presented on work [2], as a direct optimization method due to derivatives' absence. The applied algorithm estimates the source ŷn -position and designs a suitable reference γ(t)-curve, hopefully towards a sufficiently close vicinity of the actual source y ⋆ s -position. In case there are multiple critical points, the {d(y ⋆ s , ŷn )}-sequence may not converge asymptotically to a sufficiently close neighborhood of zero due to its local behavior, a challenge for these garden-like optimization algorithms. This work succeeds to obtain results in direction of the source signal position. Therefore, the proposed methodology provides an alternative for source-seeking applications by defined-to-be exploration strategy by different agent seekers, source signal maps, and obstacle modeling.