Meta-Adaptive Nonlinear Control: Theory and Algorithms

Authors: Guanya Shi, Kamyar Azizzadenesheli, Michael O'Connell, Soon-Jo Chung, Yisong Yue

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that OMAC significantly outperforms conventional adaptive control approaches which do not learn the shared representation, in inverted pendulum and 6-Do F drone control tasks under varying wind conditions1.
Researcher Affiliation Academia Caltech Purdue University {gshi,moc,sjchung,yyue}@caltech.edu, kamyar@purdue.edu
Pseudocode Yes Algorithm 1: Online Meta-Adaptive Control (OMAC) algorithm
Open Source Code Yes Code and video: https://github.com/GuanyaShi/Online-Meta-Adaptive-Control
Open Datasets No The paper describes experiments on simulated control tasks (inverted pendulum and 6-Do F quadrotor) using models, rather than relying on or providing access to publicly available datasets for training or evaluation.
Dataset Splits No The paper defines outer and inner iterations (N environments, T time steps) and uses Average Control Error (ACE) as a performance metric, but does not provide explicit train/validation/test dataset splits as it relies on simulated control environments.
Hardware Specification No The paper describes the experimental setup in terms of tasks (inverted pendulum, drone control) and controllers, but does not specify the hardware used to run these experiments (e.g., CPU/GPU models, memory).
Software Dependencies No The paper mentions algorithms and optimizers used (e.g., Adam optimizer, spectral normalization) but does not provide specific software dependencies with version numbers (e.g., 'PyTorch 1.9' or 'Python 3.8').
Experiment Setup Yes For all methods, we randomly switch the environment (wind) c every 2 s. To make a fair comparison, except no-adapt or omniscient, all methods have the same learning rate for the inner-adapter A2 and the dimensions of ˆc are also same (dim(ˆc) = 20 for the pendulum and dim(ˆc) = 30 for the drone).