Differentiable Algorithm Networks for Composable Robot Learning
This addresses the challenge of reducing data requirements and improving adaptability in robot learning systems, though it appears incremental as it builds on existing modular and end-to-end approaches.
The paper tackles the problem of robot learning by introducing Differentiable Algorithm Networks (DAN), a composable architecture that combines model-driven modular design with data-driven end-to-end learning, resulting in a simulated robot system that learns to navigate complex 3-D environments using only local visual observations and a partially correct 2-D floor map.
This paper introduces the Differentiable Algorithm Network (DAN), a composable architecture for robot learning systems. A DAN is composed of neural network modules, each encoding a differentiable robot algorithm and an associated model; and it is trained end-to-end from data. DAN combines the strengths of model-driven modular system design and data-driven end-to-end learning. The algorithms and models act as structural assumptions to reduce the data requirements for learning; end-to-end learning allows the modules to adapt to one another and compensate for imperfect models and algorithms, in order to achieve the best overall system performance. We illustrate the DAN methodology through a case study on a simulated robot system, which learns to navigate in complex 3-D environments with only local visual observations and an image of a partially correct 2-D floor map.