# Emergence, Not Bandwidth: Physical Coupling and the Limits of Learned Multi-Agent Communication

arXiv: https://arxiv.org/abs/2609.34373 · HTML: https://mlsanity.com/paper/2609.34373/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 15 of its 18 references that are in the map.
- First posted on arXiv 28 Sep 2026.

## Builds on (most cited first, 15 of 15)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Neural Discrete Representation Learning](https://mlsanity.com/paper/1711.00937/) (2017)
- [Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://mlsanity.com/paper/2108.13264/) (2021)
- [Learning to Communicate with Deep Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/1605.06676/) (2016)
- [Natural Language Does Not Emerge 'Naturally' in Multi-Agent Dialog](https://mlsanity.com/paper/1706.08502/) (2017)
- [On the Pitfalls of Measuring Emergent Communication](https://mlsanity.com/paper/1903.05168/) (2019)
- [Learning Efficient Multi-agent Communication: An Information Bottleneck Approach](https://mlsanity.com/paper/1911.06992/) (2019)
- [Anti-efficient encoding in emergent communication](https://mlsanity.com/paper/1905.12561/) (2019)
- [Limits to scalable evaluation at the frontier: LLM as Judge won't beat twice the data](https://mlsanity.com/paper/2410.13341/) (2024)
- [Biases for Emergent Communication in Multi-agent Reinforcement Learning](https://mlsanity.com/paper/1912.05676/) (2019)
- [Learning to Ground Multi-Agent Communication with Autoencoders](https://mlsanity.com/paper/2110.15349/) (2021)
- [An Analysis of Discretization Methods for Communication Learning with Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/2204.05669/) (2022)
- [Communicating Plans, Not Percepts: Scalable Multi-Agent Coordination with Embodied World Models](https://mlsanity.com/paper/2508.02912/) (2025)
- [Bandwidth-Efficient Multi-Agent Communication through Information Bottleneck and Vector Quantization](https://mlsanity.com/paper/2602.02035/) (2026)
- [Learning what to say and how precisely: Efficient Communication via Differentiable Discrete Communication Learning](https://mlsanity.com/paper/2511.01554/) (2025)

## Abstract (arXiv)

Rate-limited multi-agent teams raise three questions the emergent-communication literature has answered only empirically: what an optimal message should encode, what compression costs over a horizon, and when a learned protocol is unique enough for a teammate to read. We answer them for rate-limited Dec-POMDPs, then measure how far reinforcement learning falls short of the optimum. Our theorems fix what is achievable independently of any learner, so a gap between an engineered and a learned sender at the same bit budget is an optimization fact, not an information-theoretic one. We instantiate this on three MuJoCo arenas spanning zero, partial and rigid physical coupling, charging every condition exactly 2 bits per decision, and create the discriminating regime by closing a physical side channel within one arena, holding bodies, task and reward fixed. Communication value is governed by coupling: under rigid coupling through a shared object, no channel beats silence (+0.001 +/- 0.001, p = 0.982, n = 25), since proprioception already carries that information; without coupling, every condition solves the task; under partial coupling, the engineered 2-bit sender reaches an interquartile mean of 1.000 but the learned one reaches 0.482, indistinguishable from silence (p = 0.400, n = 25). With a shared alphabet, bandwidth cannot explain the gap. Warm-starting from an engineered receiver localizes the failure: the same channel reaches 0.857 versus 0.562 cold-started (p < 0.001), so it is neither representational nor one of maintenance; reinforcement learning fails to discover the protocol. Cross-play shows learned protocols are individually meaningful but mutually unintelligible: self-play 0.980 collapses to 0.144 across seeds, and our best constructed alignment leaves at least 77% of that gap. All headline results use 25 seeds per arena and seven published baselines at matched rate.