Summary of Simulating User Agents For Embodied Conversational-ai, by Daniel Philipov et al.
Simulating User Agents for Embodied Conversational-AI
by Daniel Philipov, Vardhan Dongre, Gokhan Tur, Dilek Hakkani-Tür
First submitted to arxiv on: 31 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Robotics (cs.RO)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel embodied agent framework is proposed to simulate user behavior in virtual environments, addressing the challenge of collecting large-scale datasets of situated human-robot dialogues. The framework uses a large language model (LLM) to simulate user goals and interactions with an embodied agent, enabling scalable and efficient dataset generation for evaluating robot interaction and task completion abilities. The LLM-based user agent is evaluated through three experiments: zero-shot prompting, few-shot prompting, and fine-tuning on the TEACh training subset. Results show that the framework achieves moderate accuracy in mimicking human speaking behavior, with an F-measure of 42% using zero-shot prompting and 43.4% using few-shot prompting. Fine-tuning improves the performance in deciding what to say from 51.1% to 62.5%. This approach has implications for research in reinforcement learning using AI feedback. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about building a computer program that can help people use robots more easily. The program uses a special kind of artificial intelligence called a large language model. This allows it to understand and follow instructions, as well as communicate with the robot. The goal is to make collecting data for this type of interaction easier and faster. The paper tests this approach by comparing how well the program can mimic human-like behavior when interacting with a robot. The results show that it’s possible to use this method to assess and improve the way robots complete tasks using natural language communication. |
Keywords
» Artificial intelligence » Few shot » Fine tuning » Large language model » Prompting » Reinforcement learning » Zero shot
