Generalize Dimensionality of Monty's Computations
We are developing LMs that support 1D and 2D reference frames, but our computations all assume more expensive 3D representations.
Recent work with sensor modules has enabled reference frames that model 2D space, and we hope to make a similar advance with 1D space. From a computational efficiency perspective however, the LMs that receive these inputs still assume that space is 3D. This restricts our ability to see greater computational benefits from using low-dimension representations. This Future Work item is therefore about refactoring Monty's representations, both those sent between LMs, and the reference frames used within LMs, to be of lower dimensionality where this is consistent with the sensory input they receive.
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 4 days ago