CoDiMAD: Diffusion-Based Privileged Distillation for Communication-Free Multi-Robot Coordination
2026-07-10 • Robotics
Robotics
AI summaryⓘ
The authors address the problem of coordinating multiple robots when each robot only has limited, local information and cannot communicate with others. They note that traditional methods that try to mimic a smart, fully informed controller often produce weak or confused actions because the same local view can mean different things globally. To fix this, they propose CoDiMAD, which uses a special model called a diffusion probabilistic model to better capture the range of good actions a robot can take. Their approach outperforms existing methods in tasks requiring teamwork without communication. They also provide theoretical explanations for why their method works better.
multi-robot coordinationpartial observabilityprivileged policy distillationmulti-modal action distributiondeterministic distillationconditional denoising diffusion probabilistic modelsMAPPOoffline datasetmulti-agent reinforcement learning (MARL)
Authors
Jiyue Tao, Shunheng Xin, Tongsheng Shen, Dexin Zhao, Feitian Zhang
Abstract
Decentralized multi-robot coordination under partial observability remains challenging, especially in communication-free settings where agents must act solely from local sensor observations. Privileged policy distillation provides a promising approach by transferring knowledge from a globally informed oracle to sensor-constrained students. However, in multi-agent systems, the same local observation may correspond to multiple global configurations requiring qualitatively different cooperative actions, making the conditional action distribution inherently multi-modal. Standard deterministic distillation collapses these modes to their mean, often yielding invalid or hesitant actions. To address this issue, we propose CoDiMAD, a three-stage framework that trains a privileged oracle with MAPPO, constructs an offline dataset of local-observation-oracle-action pairs, and distills the oracle into decentralized students parameterized as conditional denoising diffusion probabilistic models. By approximating the conditional oracle-action distribution through the diffusion reverse process, CoDiMAD samples decisive actions from coherent coordination modes rather than averaging across them. Theoretical analysis characterizes the mode-averaging failure of deterministic distillation and the distributional recovery property of diffusion-based distillation. Experiments on three cooperative tasks show that CoDiMAD consistently outperforms direct local MARL and deterministic distillation baselines. The source code will be made publicly available upon acceptance.