Get Started
Home
Topics
Search
Library
Research questionHow can multilayer RNNs learn nonlinear quadrotor dynamics without vanishing or exploding gradients?Deep multilayer RNNs must propagate learning signals through both time and network depth, making it difficult to capture high-order nonlinear dynamics reliably. This is especially problematic when learning a quadrotor’s altitude behavior from data.
AI
Machine Learning
Neural and Evolutionary Computing
Robotics
Latest papersRecent research connected to this question, newest first.Modular Deep Recurrent Neural Network: Application to QuadrotorsThe source presents a modular deep RNN framework with automatic derivative computation and evaluates a feedforward-connected multilayer architecture on learned quadrotor altitude dynamics. It reports improved learning of high-order dynamics and nonlinearities, along with alleviated spatial vanishing or exploding gradients; the evidence is limited to the described quadrotor case study and comparisons with existing methods.research paper · Sep 3, 2026
Related questions
How can deep continuous-time recurrent networks avoid depth-induced gradient attenuation under truncated temporal backpropagation?How can we learn an effective LQR controller from unknown dynamics without a stable initial policy?How does gradient training build hierarchical representations beyond the lazy kernel regime?When do explicitly organized recurrent interactions improve dynamical learning over generic reservoirs under matched state dimensions and controlled tuning?