The research paper focuses on the development of a Hierarchical Meta-learning-based Adaptive Controller (HMAC), which enables accurate and swift online adaptation in fluctuating environments. HMAC is developed to learn from and adapt to multiple sources of disturbances, particularly in the context of robotics. The disturbances are categorized into manageable ones, which can be directly monitored or controlled during training, and latent ones, which cannot be directly measured or controlled. The HMAC approach has shown better performance in adaptation to these multi-source disturbances compared to other adaptive controllers.
Publication date: 22 Nov 2023
Project Page: https://sites.google.com/view/hmacproject
Paper: https://arxiv.org/pdf/2311.12367
