The field of artificial intelligence is witnessing significant advancements in the development of vision-language models for robotic learning and exploration. These models are being designed to improve the ability of robots to understand and interact with their environment, and to learn from their experiences. One of the key areas of focus is the development of memory systems that can efficiently store and retrieve information, allowing robots to learn from their experiences and adapt to new situations. Another important area of research is the development of frameworks that can jointly model and enhance object detection and relationship classification in open-vocabulary scenarios, enabling robots to better understand their environment and make more informed decisions. Furthermore, researchers are exploring the use of vision-language models to generate high-level exploratory behaviors, allowing robots to explore their environment in a more efficient and effective manner. Notable papers include:

• A Grounded Memory System For Smart Personal Assistants, which proposes a memory system consisting of three components to efficiently manage relational information.
• METOR: A Unified Framework for Mutual Enhancement of Objects and Relationships, which achieves state-of-the-art performance in open-vocabulary video visual relationship detection.
• Imagine, Verify, Execute: Memory-Guided Agentic Exploration with Vision-Language Models, which presents a framework for agentic exploration that leverages vision-language models to abstract RGB-D observations into semantic scene graphs and generate executable skill sequences.