Learning to Communicate Through Implicit Communication Channels
Han Wang, Binbin Chen, Tieying Zhang, Baoxiang Wang
TL;DR
This paper introduces the Implicit Channel Protocol (ICP), a novel framework enabling agents to communicate implicitly through specialized actions, improving information transmission efficiency in multi-agent tasks without relying on theory of mind.
Contribution
The paper proposes the ICP framework that encodes messages via scouting actions and develops training algorithms, advancing implicit communication methods beyond ToM-based approaches.
Findings
ICP outperforms baseline methods in Guessing Numbers, Revealing Goals, and Hanabi.
ICP enables more efficient information transmission in multi-agent tasks.
Training algorithms effectively learn messaging and acting strategies.
Abstract
Effective communication is an essential component in collaborative multi-agent systems. Situations where explicit messaging is not feasible have been common in human society throughout history, which motivate the study of implicit communication. Previous works on learning implicit communication mostly rely on theory of mind (ToM), where agents infer the mental states and intentions of others by interpreting their actions. However, ToM-based methods become less effective in making accurate inferences in complex tasks. In this work, we propose the Implicit Channel Protocol (ICP) framework, which allows agents to communicate through implicit communication channels similar to the explicit ones. ICP leverages a subset of actions, denoted as the scouting actions, and a mapping between information and these scouting actions that encodes and decodes the messages. We propose training algorithms…
Peer Reviews
Decision·ICLR 2025 Poster
The paper is well-structured and easy to follow. The algorithm is clearly-explained, with clear definitions of necessary notations. The experiments are well-designed and comprehensive, with sufficient implementation details.
1. The ICP framework requires pre-identified scouting actions that can serve as indirect communication channels. This dependency limits its applicability in environments where such actions are not readily available or are difficult to define. 2. It can help the readers to better understand the method if the authors can include a diagram of the algorithm pipeline.
The proposed method is interesting, particularly toward the foundational problem of efficient multi agent communication by generating a information table with respect to the observation and message. Additionally, research into reducing computational complexity for intention inference techniques of agents will be highly valuable to the large-scale multi-agent systems.
1. The paper does not discuss the impact of scalability or impact of heterogeneous agents to the proposed framework for multi agent systems, as you increase the number of agents to say up to 10, 20, 50 agents. 2. The paper does not compare results for common communication architectures such as CommNet, TarMAC, SARNet etc for their environments but only with value decomposition networks (VDN), that do not perform communication as part of their actions. 3. The paper does not discuss limitations
- The effectiveness of the method is evaluated against two newly designed benchmarks and results show ICP being superior to baselines
- I fail to comprehend how the framework is a form of implicit communication. My understanding of implicit communication is the use of environment actions to communicate information (e.g., learning a protocol such that walking forward means yes and walking backward means no). ICP proposes add actions to the action space that maps information into certain ‘embeddings’ to be broadcasted to other agents. How is this not explicit communication? - Unless I am missing something, the channels are prov
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsInnovative Teaching and Learning Methods