Neural Hebbian Plastic Control Network for Adaptive Locomotion

Neural synaptic plasticity is fundamental to adaptive capabilities in animals. Inspired by this, researchers have developed robotic controllers with plastic neural networks, employing Hebbian learning rules to achieve adaptive locomotion in unforeseen situations during training. Previous studies demonstrated that a neural network with Hebbian learning can adapt to morphological damage in a simulated quadrupedal robot and physical complex-legged robots, even when the damage was not seen during training. Although the Hebbian network enables adaptation to unseen situations in locomotion tasks, it remains unclear whether it learns high-level inter-leg coordination or merely encodes specific joint trajectories for adaptation. To address this question, this study investigates a Hebbian control network by systematically shuffling the legs between the left and right sides analyzing their impact on locomotor adaptation. To this end, the study can pave the way for the locomotion control development of modular and reconfigurable robots with plug-and-play limbs, enabling functionality without relying on predefined limb configurations.