Reproduce “Efficient Deep Learning of Robust Policies from MPC using Imitation and Tube-Guided Data Augmentation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Andrea Tagliabue, Jonathan P. How
A new Imitation Learning strategy, Sampling Augmentation (SA), efficiently trains robust deep neural network policies by distilling knowledge from Model Predictive Control (MPC) experts. SA enables zero-shot sim-to-real and lab-to-real transfer with 100% success rates on complex tasks using as few as one demonstration, significantly outperforming baselines in data and training efficiency.
Key insights
- The inherent robustness properties of RTMPC, particularly its disturbance-invariant sets (tubes), can be effectively exploited to guide data augmentation for learning robust policies.
- Sampling within the RTMPC tube allows for efficient generation of diverse training data that covers states likely to be encountered under bounded uncertainties, significantly improving robustness and sample efficiency.
- Parametric sensitivity analysis of the ancillary Nonlinear MPC (NMPC) in nonlinear RTMPC enables computationally efficient data augmentation without solving the full NMPC for each augmented sample.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.