Stanford-ILIAD
Inquiry About Your Research on Explore Embodied QA
I have a few questions regarding some specific aspects of your work and would greatly appreciate your clarification.
-
Relationship between Explored and Unoccupied Regions: Could you please elaborate on the relationship between the explored and unoccupied regions? Why does the unoccupied region appear significantly larger than the explored region? Isn't the explored region the sum of the known occupied and unoccupied areas?
-
Orientation Adjustment: Your paper emphasizes path planning in terms of spatial position changes. However, could you provide more details on how the robot's orientation is adjusted during this process?
-
Initial Exploration Circle: In the initial step, a circular area around the robot is already explored. Could you explain the rationale behind this initial explored area?
-
Fine-Tuning of VLM: The paper mentions that the first 300 questions are used for fine-tuning the VLM. Was this fine-tuning conducted using the LLAVA method for visual question answering?
