The development of bio-inspired robots is limited by the challenge of controlling complex morphologies with nonlinear, soft actuation. Current robots often rely on analytical models of joint dynamics, while learning-based approaches require domain-specific tuning, limiting generalisation to other robots. This project aims to create a scalable, generically applicable framework for designing and optimising distributed controllers for bio-inspired robots. We propose a graph neural network control system, where processing power is embedded in graph nodes, automatically structured to match the robot’s morphology. This modular approach enables scalability and reduces compatibility issues between control and body dynamics. Our system leverages shared policies to improve skill reuse while allowing for specialised node behaviour, driven by either memory mechanisms (gated recurrent units) or neuroplasticity-based self-organisation. The distributed architecture and plasticity mechanisms also enhance robustness and adaptability. We hypothesise that (1) centralised intelligence will emerge through message passing, coordinating node specialisation, and that (2) Hebbian-inspired local learning can optimise the stability-plasticity trade-off, enabling experience-driven online adaptation. By introducing modularity and plasticity into artificial agents, this research has the potential to accelerate the development of the next generation of adaptive, autonomous, and safe robotic systems.