The field of multi-modal control and referring expression segmentation is witnessing significant developments, with a focus on enhancing the flexibility and accuracy of models. Researchers are exploring innovative approaches to adapt behavior foundation models to specific tasks, improving their performance while preserving their generalization capabilities. Moreover, there is a growing emphasis on fine-grained cross-modal alignment and decoding, enabling more effective integration of visual and textual features. Noteworthy papers include:

• Task Tokens, which introduces a method to tailor behavior foundation models to specific tasks through reinforcement learning, and
• CADFormer, which proposes a fine-grained cross-modal alignment and decoding Transformer for referring remote sensing image segmentation. These advancements have the potential to significantly impact various applications, from humanoid control to remote sensing image analysis.