LOC-ZSON: Language-driven Object-Centric Zero-Shot Object Retrieval and Navigation

TL;DR

LOC-ZSON introduces a language-driven, object-centric image representation and training approach that enhances zero-shot object retrieval and navigation in complex scenes, demonstrating significant improvements in success rates across simulated and real-world environments.

Contribution

The paper presents a novel object-centric image representation and a training framework using language models for improved zero-shot object navigation and retrieval.

Findings

Achieves 1.38-13.38% improvement in text-to-image recall.

Shows 5% and 16.67% better success rates in simulation and real-world navigation.

Introduces LLM-based augmentation and prompt templates for stable training.

Abstract

In this paper, we present LOC-ZSON, a novel Language-driven Object-Centric image representation for object navigation task within complex scenes. We propose an object-centric image representation and corresponding losses for visual-language model (VLM) fine-tuning, which can handle complex object-level queries. In addition, we design a novel LLM-based augmentation and prompt templates for stability during training and zero-shot inference. We implement our method on Astro robot and deploy it in both simulated and real-world environments for zero-shot object navigation. We show that our proposed method can achieve an improvement of 1.38 - 13.38% in terms of text-to-image recall on different benchmark settings for the retrieval task. For object navigation, we show the benefit of our approach in simulation and real world, showing 5% and 16.67% improvement in terms of navigation success rate, respectively.