Mechatronics for noise control: on the capabilities of ENC

Evolving neuromorphic flight control for a flapping‐wing mechanical insect

“…Larger neuron counts are easily possible; alternatively, multiple chips can be interconnected to increase neuron counts. Our chip supports both intrinsic and extrinsic evolution equally well Vigraham and Gallagher, 2006). For most successful micro robot applications based on evolutionary techniques, especially for flapping-wing bots, off-line model-based learning would be preferred.…”

“…The delay networks are placed to produce asymmetric/symmetric actuations of left wing with respect to right wing or vice-a-versa to produce net non-zero/zero torques and forces in Y-direction or for changing the overall orientation of the robot (if required). Based on the evolutionary approach applied in and Vigraham and Gallagher (2006), we have evolved a gating network whose output behavior is similar to that of a analog multiplier with input signals being actuation signal (stroke or rotation) and a delay input signal. For the two set of experiments, autonomous, and non-autonomous controller evolutions, that would be described in this paper, a fully connected eight neuron network with one senor per neuron is chosen for the central core network block and the evolutionary algorithm is employed to evolve the central core network and/or the delay duration (duration for which external control is high) for the pre-evolved outer delay networks.…”

“…The neuromorphic evolvable hardware, also referred as continuous time recurrent neural network‐evolvable hardware (CTRNN‐EH) architecture, introduced in Boddhu et al (2008), Gallagher (2003) and Vigraham and Gallagher (2006) is summarized schematically in Figure 1. The CTRNN‐EH architecture is a neuromorphic variant of the standard evolvable hardware paradigm using continuous time recurrent neural networks (CTRNNs) as the reconfigurable hardware substrate.…”

“…The very‐large‐scale integration (VLSI) chip‐CTRNN shown in Figure 2, is based on the current‐mode integrated circuit designs (Gallagher et al , 2007; Vigraham and Gallagher, 2006) and was fabricated using 0.6 μ m complementary metal‐oxide‐semiconductor technology. The design in its present form can support up to a fully connected four‐neuron network.…”

“…In recent years, the authors and their colleagues have fielded evolved neuromorphic controllers for legged robots. These controllers were demonstrated to pack a large amount of functionality into very compact analog circuitry (Brown, 2005; Vigraham and Gallagher, 2006; Gallagher et al , 2007). The neuromorphic evolvable hardware method employed for the legged robot controller study has been effective to exploit and accommodate the variations and dynamics of the robot, that were not previously conceived completely in the human designed controllers for the same (Gallagher et al , 1996; Chiel et al , 1999; Beer et al , 1999).…”