120 Minutes and a Laptop: Minimalist Image-goal Navigation via Unsupervised Exploration and Offline RL
This work addresses the challenge of high computational and data requirements for robotic navigation, making it more accessible for rapid prototyping, though it is incremental in combining existing techniques like unsupervised exploration and offline RL.
The paper tackles the problem of image-goal visual navigation by proposing a method that enables dataset collection, training, and real-world deployment in under 120 minutes on a consumer laptop without human intervention, showing improved exploration efficiency and outperforming zero-shot baselines.
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.