Modular Deep Recurrent Neural Network: Application to Quadrotors
2026-09-07 12:00Models🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
0mentions
SummaryAI generated
The researchers proposed a new type of modular deep recursive neural network (RNN) that automatically calculates gradients by adding connections between feedforward layers, effectively solving the problem of gradient vanishing and explosion encountered by traditional RNNs in multi-layer structures. This architecture significantly enhances the model’s ability to capture high-order dynamics and nonlinear relationships. The research team successfully built the dynamics model of quadcopter aircraft using this network, verifying its generalizability in complex aircraft control scenarios and overcoming the limitations of existing methods, which struggle to adapt quickly or achieve full generalization.
The researchers introduced a modular deep recursive neural network (RNN) aimed at facilitating the deployment of various RNN architectures and automatically calculating gradients. By adding inter-layer feedforward connections in multi-layer RNNs, the model’s ability to learn high-order dynamics and nonlinear relationships was significantly improved, while also alleviating the problem of gradient vanishing/explosion in space. This achievement was verified in the case of quadcopters: this network structure successfully learned highly dynamic models, whereas existing methods could not generalize quickly or at all.