Decentralized Safe and Scalable Multi-Agent Control under Limited Actuation

TL;DR

This paper tackles safe decentralized multi-agent control under actuation limits by introducing MA-ICBF, a two-stage framework that first learns Neural Integral Control Barrier Functions (NICBF) and a decentralized safe policy, then augments the policy with a lightweight MPC-ICBF to guarantee safety and input constraint satisfaction. It also proposes a gradient-inspired deadlock-minimization mechanism and analyzes the optimization structure to escape local minima. Empirical results show MA-ICBF achieves 100% collision avoidance across environments and scales to over 1000 agents, with substantial deadlock reductions (up to ~92%) and strong adherence to input bounds, while offering computational benefits via a log-sum-exp-based MPC constraint aggregation. The method generalizes across different agent counts, enabling safe, scalable real-time control in cluttered settings, though it assumes known dynamics and invites future work in vision-based and dynamic-obstacle scenarios.

Abstract

To deploy safe and agile robots in cluttered environments, there is a need to develop fully decentralized controllers that guarantee safety, respect actuation limits, prevent deadlocks, and scale to thousands of agents. Current approaches fall short of meeting all these goals: optimization-based methods ensure safety but lack scalability, while learning-based methods scale but do not guarantee safety. We propose a novel algorithm to achieve safe and scalable control for multiple agents under limited actuation. Specifically, our approach includes: $(i)$ learning a decentralized neural Integral Control Barrier function (neural ICBF) for scalable, input-constrained control, $(ii)$ embedding a lightweight decentralized Model Predictive Control-based Integral Control Barrier Function (MPC-ICBF) into the neural network policy to ensure safety while maintaining scalability, and $(iii)$ introducing a novel method to minimize deadlocks based on gradient-based optimization techniques from machine learning to address local minima in deadlocks. Our numerical simulations show that this approach outperforms state-of-the-art multi-agent control algorithms in terms of safety, input constraint satisfaction, and minimizing deadlocks. Additionally, we demonstrate strong generalization across scenarios with varying agent counts, scaling up to 1000 agents.

Figures (8)

• Figure 1: Demonstrating our decentralized safe and scalable multi-agent control in two environments. Blue and yellow denote goals and current positions, respectively.
• Figure 2: Training
• Figure 3: Execution
• Figure 5: Empty
• Figure 6: Maze