1. Find and Deliver:

Based on instructions and descriptions, locate the designated item and deliver it to a designated individual. The model or system should understand context, infer possible locations, and accommodate regular human behavior.
Task-specific: including obtaining photos, introductions, and approximate schedules of characters. Success conditions involve retrieving items and identifying the target person within close proximity.
Evaluation: based on the total time taken and success rate.

2. Device Operation Instruction Following:

Operate device components as per manual and guidance, interacting with the device to fulfill its functions. This task tests multimodal models' ability to align language with vision, mapping semantic concepts onto physical functions.
Task Setting: interacting with devices in a specified order to execute described functions.
Test benchmark: time taken, number of attempts, and success rate.

3. Device Manipulation Summary:

Exploration with the device, observe feedback, and answer questions based on experience. This aims to transform personal experiences into usable knowledge, exploring the model's behavior understanding and knowledge extraction.
Task setting: generating interactive devices that provide feedback on specific operations.
Evaluation: answering questions (multiple-choice) and summarizing the operation manual (fill-in-the-blank) within the specified time frame. Metrics will be based on answer accuracy and time taken.

4. Requirement Following:

Provide personalized services to diverse human NPCs, meeting their basic needs, offering assistance, and detecting changes in NPC status post-execution to align with their preferences and personalized requirements.

To Be Announced !