Differentiable Programming and Neural ODEs for Accelerating Model Based Reinforcement Learning and Optimal Control

Paul Shen
Paul Shen
Sep 3, 2020 · 10 min read
Strategies learnt under a minute: 1-go swing up (left), resonant incremental swing up with force constraint (right)

We will explain the theory in detail first. Feel free to jump to the code section.

We simplify and accelerate training in model based reinforcement learning problems by using end-to-end differentiable programming in Julia. We compute…