Can we change the CMP to use displacements instead of locations?
Investigate using movement vectors instead of locations in the CMP and the implications it would have for Monty.
Although there is reasonable evidence that the brain may be using movements instead of locations, it is unclear whether there are any computational benefits to doing this. As such, the
improved-metricfield is left open. A first step to tackling this task would be to think through any functional implications of this change. In the meantime, we include this task because choosing the right assumptions about a system is key to developing long-term, successful solutions. Using displacements rather than locations affects how both voting and motor actions can be implemented. Continuing to use locations, and diverging from a more brain-like approach, may therefore lead us into a "local minimum" style solution for some elements in Monty.
Movement is core to how LMs process and model the world. Currently, an LM receives an observation encoded with a body-centric location, and then infers a displacement in object-centric coordinates. Similarly, goals are specified as a target location in body-centric coordinates, which are then acted upon.
However, a more general formulation might be to use displacements as the core spatial information in the CMP, such that a specific location (in body-centric coordinates or otherwise) is not the primary form of communication outside of an LM or sensor module.
Such an approach might align well with adding information about flow (see Detect Local and Global Flow), modeling moving objects (see Deal With Moving Objects), and supporting abstract movements like the transition from grandchild to grandparent. It would also result in a reformulation of "goals" to "goal-displacements".
While this Future Work item could be viewed as falling in the category of CMP/hierarchy improvements, the most significant obstacle for this change is the requirement to reformulate voting such that a shared coordinate system with locations is not necessary. A voting algorithm of this kind remains a theoretical gap, although we discuss recent ideas in this video.
Help Us Make This Page Better
All our docs are open-source. If something is wrong or unclear, submit a PR to fix it!
Updated 26 days ago