Real2Sim Transfer using Differentiable Physics

Accurate simulations allow modern machine learning techniques to be applied to robotics problems, with samplecollection runtimes orders of magnitudes faster than the real world. Current reinforcement learning approaches require laborious manual calibration of carefully designed models, or, in a model-free context, vast amounts of training data to acquire such accurate models from real-world trials. In this work, we introduce a new layer in the deep learning toolbox that imposes a strong inductive bias to generate physically accurate predictions of rigid-body dynamics and allows for the automatic inference of system parameters given an ad-hoc model description.