Dynamics of behavior: Theory and applications for autonomous robot architectures

摘要: Abstract Limitations both for the further development as well actual technical application of autonomous robots arise from lack a unifying theoretical language. We propose three concepts such language: (1) Behaviors are represented by variables, specific constant values which correspond to task demands; (2) generated attractors dynamical systems; (3) Neural field dynamics lift these dynamic principles representation information. show how can be used design robots. Because behaviors attractor states systems, robot architecture addresses control-theoretic stability. Moreover, flexibility arises bifurcations in behavioral dynamics. Therefore techniques qualitative theory systems and tune architectures. demonstrate ideas two implementations. In one case, visual sensory information is integrated achieve target acquisition obstacle avoidance an vehicle minimizing known problem spurious states. second implementation same behavior, neural endows system with form memory. A critical discussion approach highlights strengths weaknesses compares other efforts this direction.