Deep-Unfolded Coordination

2026-06-18

AI Summary

This work addresses the challenges of tedious manual parameter tuning and poor generalization in multi-agent distributed optimization by proposing the Deep Coordinator framework. It introduces, for the first time, a deep unfolding of the non-convex ADMM-DDP solver into a neural network architecture, wherein learnable modules dynamically adjust penalty parameters during optimization to enable adaptive convergence. An unsupervised training strategy is incorporated to prevent degenerate solutions, substantially enhancing computational efficiency. Experimental results on vehicle platooning and quadrotor simulations demonstrate that, while maintaining comparable trajectory quality, the proposed method achieves a 6.18–9.44× speedup in solution time and retains its performance advantage even when the system scale is increased eightfold.

Abstract

Distributed optimization is a highly scalable and structurally transparent technique to solve multi-agent robotics problems; however, such methods often suffer from the need for highly-specialized, problem-specific hyperparameter tunings. In this work, we propose Deep Coordinator, a deep-unfolding framework that learns to dynamically adjust the hyperparameters of ADMM-DDP, a popular distributed solver for robotics tasks, at solve-time in response to optimizer performance. Our architecture consists of unrolling a fixed number of ADMM-DDP iterations into a neural network with learnable functions between layers mapping the optimizer state to the next hyperparameters. To the best of our knowledge, Deep Coordinator is the first deep-unfolding framework to adapt the penalty parameters of a non-convex optimizer at solve-time; we show that the mainstream supervised approach can yield degenerate solutions when training such models, and propose an unsupervised learning scheme. On simulations with fleets of cars and quadrotors, Deep Coordinator produces trajectories of comparable quality 6.18-9.44x faster than conventional solvers. Furthermore, Deep Coordinator retains its performance benefits when deployed to systems up to 8x larger than trained on.

Problem

• distributed optimization
• hyperparameter tuning
• multi-agent robotics
• ADMM-DDP
• non-convex optimization

Innovation

• deep unfolding
• distributed optimization
• ADMM-DDP
• hyperparameter adaptation
• unsupervised learning