Meta Adaptation using Importance Weighted Demonstrations

TL;DR

This paper introduces a meta-learning algorithm that uses importance weighting of past demonstrations to enable robots to adapt quickly to new, unseen tasks in dynamic environments, improving sample efficiency and generalization.

Contribution

The authors propose a novel importance-weighted demonstration method for meta imitation learning, enhancing adaptation to new tasks with limited data in changing environments.

Findings

Effective adaptation to unseen environments demonstrated

Improved sample efficiency in task generalization

Successful real-world robot navigation experiments

Abstract

Imitation learning has gained immense popularity because of its high sample-efficiency. However, in real-world scenarios, where the trajectory distribution of most of the tasks dynamically shifts, model fitting on continuously aggregated data alone would be futile. In some cases, the distribution shifts, so much, that it is difficult for an agent to infer the new task. We propose a novel algorithm to generalize on any related task by leveraging prior knowledge on a set of specific tasks, which involves assigning importance weights to each past demonstration. We show experiments where the robot is trained from a diversity of environmental tasks and is also able to adapt to an unseen environment, using few-shot learning. We also developed a prototype robot system to test our approach on the task of visual navigation, and experimental results obtained were able to confirm these suppositions.