First spiking neural controller coordinates humanoid walking and arm reaching

Spiking Neural Networks (SNNs) promise ultra-low-power control for humanoid robots when run on neuromorphic chips, but prior work kept locomotion and arm control separate. Now, researchers at the University of Waterloo have built the first spiking architecture that coordinates both in a full-scale simulated humanoid. Their system uses the Neural Engineering Framework (NEF) and Semantic Pointer Architecture (SPA) to represent and transform neural populations, while a biologically grounded spiking basal ganglia model decides when to walk or reach.

The team validated their approach through co-simulation of Nengo (for neural control) and NVIDIA Isaac Sim (for physics). The humanoid successfully reached targets, drew continuous digits on a surface, followed a path while walking, and switched between locomotion and arm control—all driven purely by spikes. This integration is crucial for real-world robots that must navigate and manipulate objects simultaneously, and since the controller is fully spike-based, it can be directly deployed on energy-efficient neuromorphic hardware like Intel's Loihi or BrainScaleS.