MAGIK: Mapping to Analogous Goals via Imagination-enabled Knowledge Transfer

TL;DR

MAGIK introduces an imagination-based framework enabling reinforcement learning agents to perform zero-shot transfer to new tasks by mapping entities to analogous ones, reducing retraining needs.

Contribution

It presents a novel method for knowledge transfer in RL using imagination to map entities across tasks without environment interaction.

Findings

Effective zero-shot transfer on MiniGrid and MuJoCo tasks.

Outperforms baseline methods in transfer scenarios.

Requires only a small number of human-labelled examples.

Abstract

Humans excel at analogical reasoning - applying knowledge from one task to a related one with minimal relearning. In contrast, reinforcement learning (RL) agents typically require extensive retraining even when new tasks share structural similarities with previously learned ones. In this work, we propose MAGIK, a novel framework that enables RL agents to transfer knowledge to analogous tasks without interacting with the target environment. Our approach leverages an imagination mechanism to map entities in the target task to their analogues in the source domain, allowing the agent to reuse its original policy. Experiments on custom MiniGrid and MuJoCo tasks show that MAGIK achieves effective zero-shot transfer using only a small number of human-labelled examples. We compare our approach to related baselines and highlight how it offers a novel and effective mechanism for knowledge transfer via imagination-based analogy mapping.