We propose a planning system suitable for: • daily living support for people with mental disease encompassing the use of tens or hundreds of objects • humanoid robots with advanced sensory-motor skills The domain representation is based on concepts related to affordance and capability: • each object or location has affordances indicating possibilities for action • an agent must have all the relevant capabilities to respond to affordances A planning algorithm using conceptual spaces for representing both affordances and capabilities is introduced, which adopts techniques borrowed from neural networks and self-organizing maps.