TL;DR of the dimensions. Please check out the taxonomy for more details.

• Text: non-embodied agents perceive and act with texts, e.g., chatbots
• virtual agents: non-embodied agents perceive and act within the virutal environment, e.g., web agents
• embodied agents: agents that interact with the world through the observation and actions of "bodies". e.g., simulated robots/human characters.
• robotics: real world embodied agents.

• Note: Here we mainly focus on building agents with neural models.

• Qualitative: Indicate the evaluation is based on some examples.
• Model-based: Use neural model for end-to-end evaluation.

• mixed_objectives: A mixture of collaboration, competition.
• implicit_objectives: No goal/task mentioned explicitly for the social agents (e.g., chit-chat agent; the implied objective is to make users engage.)

• human_agent: Real humans involved in the interactions.