Few-shot model-based adaptation in noisy conditions

TL;DR

This paper introduces an uncertainty-aware Kalman filter-based neural network for few-shot adaptation of dynamics models in noisy environments, enhancing transfer accuracy in robotics applications.

Contribution

It presents a novel Kalman filter-inspired neural network architecture explicitly designed to handle domain noise during few-shot adaptation in physical systems.

Findings

Improves adaptation error over baseline LSTM models

Outperforms model-free reinforcement learning in noisy conditions

Enables system analysis through hidden state examination

Abstract

Few-shot adaptation is a challenging problem in the context of simulation-to-real transfer in robotics, requiring safe and informative data collection. In physical systems, additional challenge may be posed by domain noise, which is present in virtually all real-world applications. In this paper, we propose to perform few-shot adaptation of dynamics models in noisy conditions using an uncertainty-aware Kalman filter-based neural network architecture. We show that the proposed method, which explicitly addresses domain noise, improves few-shot adaptation error over a blackbox adaptation LSTM baseline, and over a model-free on-policy reinforcement learning approach, which tries to learn an adaptable and informative policy at the same time. The proposed method also allows for system analysis by analyzing hidden states of the model during and after adaptation.