Abstract

Typically, AI researchers and roboticists try to realize intelligent behaviorin machines by tuning parameters of a predefined structure (body plan and/orneural network architecture) using evolutionary or learning algorithms. Anotherbut not unrelated longstanding property of these systems is their brittlenessto slight aberrations, as highlighted by the growing deep learning literatureon adversarial examples. Here we show robustness can be achieved by evolvingthe geometry of soft robots, their control systems, and how their materialproperties develop in response to one particular interoceptive stimulus(engineering stress) during their lifetimes. By doing so we realized robotsthat were equally fit but more robust to extreme material defects (such asmight occur during fabrication or by damage thereafter) than robots that didnot develop during their lifetimes, or developed in response to a differentinteroceptive stimulus (pressure). This suggests that the interplay betweenchanges in the containing systems of agents (body plan and/or neuralarchitecture) at different temporal scales (evolutionary and developmental)along different modalities (geometry, material properties, synaptic weights)and in response to different signals (interoceptive and external perception)all dictate those agents' abilities to evolve or learn capable and robuststrategies.

 

Introduction (beta)

None

 

Conclusion (beta)

None