Vector trace cells (VTC) suggest a vector-based model of computing spatial relationships between an agent and multiple spatial objects, or between different objects.
There is this paper from Deepmind where boundary vector cells (BVCs), and place cells, as well as egocentric boundary vector cells (egoBVCs) emerged but not VTC.
Object = Combinaison (features + vector between features).
Action = Combinaison (objects/features + vector between objects/features).
This should allow a lot more robust representations and faster generalisation.