Mobile Manipulation with Active Inference for Long-Horizon Rearrangement Tasks

TL;DR

This paper presents a hierarchical active inference framework for mobile manipulation that enables flexible, goal-directed long-horizon tasks in realistic robotic environments, outperforming existing methods without offline training.

Contribution

It introduces a novel hierarchical active inference architecture that combines high-level skill selection with low-level control, scalable to complex robotic benchmarks.

Findings

Outperforms state-of-the-art baselines on Habitat Benchmark tasks

Enables online adaptability and recovery from failures

Scales active inference to complex, real-world robotic tasks

Abstract

Despite growing interest in active inference for robotic control, its application to complex, long-horizon tasks remains untested. We address this gap by introducing a fully hierarchical active inference architecture for goal-directed behavior in realistic robotic settings. Our model combines a high-level active inference model that selects among discrete skills realized via a whole-body active inference controller. This unified approach enables flexible skill composition, online adaptability, and recovery from task failures without requiring offline training. Evaluated on the Habitat Benchmark for mobile manipulation, our method outperforms state-of-the-art baselines across the three long-horizon tasks, demonstrating for the first time that active inference can scale to the complexity of modern robotics benchmarks.