120 Minutes and a Laptop: Minimalist Image-goal Navigation via Unsupervised Exploration and Offline RL

TL;DR

This paper introduces MINav, a minimalist approach to image-goal navigation that trains and deploys a policy in under 120 minutes on a consumer laptop without human intervention, using offline reinforcement learning.

Contribution

MINav demonstrates that effective image-goal navigation can be achieved with minimal data collection and computational resources through offline RL and unsupervised exploration.

Findings

MINav outperforms zero-shot baselines in target environments.

Training and deployment are completed in less than 120 minutes.

The approach scales well with dataset size and requires no human intervention.

Abstract

The prevailing paradigm for image-goal visual navigation often assumes access to large-scale datasets, substantial pretraining, and significant computational resources. In this work, we challenge this assumption. We show that we can collect a dataset, train an in-domain policy, and deploy it to the real world (1) in less than 120 minutes, (2) on a consumer laptop, (3) without any human intervention. Our method, MINav, formulates image-goal navigation as an offline goal-conditioned reinforcement learning problem, combining unsupervised data collection with hindsight goal relabeling and offline policy learning. Experiments in simulation and the real world show that MINav improves exploration efficiency, outperforms zero-shot navigation baselines in target environments, and scales favorably with dataset size. These results suggest that effective real-world robotic learning can be achieved with high computational efficiency, lowering the barrier to rapid policy prototyping and deployment.