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. Videos andcode are available at https://maicbf.github.io/.

In this paper, we address the problem of reducing the computational burden of Model Predictive Control (MPC) for real-time robotic applications. We propose TransformerMPC, a method that enhances the computational efficiency of MPC algorithms by leveraging the attention mechanism in transformers for both online constraint removal and better warm start initialization. Specifically, TransformerMPC accelerates the computation of optimal control inputs by selecting only the active constraints to be included in the MPC problem, while simultaneously providing a warm start to the optimization process. This approach ensures that the original constraints are satisfied at optimality. TransformerMPC is designed to be seamlessly integrated with any solver, irrespective of its implementation. To guarantee constraint satisfaction after removing inactive constraints, we perform an offline verification to ensure that the optimal control inputs generated by the solver meet all constraints. The effectiveness of TransformerMPC is demonstrated through extensive numerical simulations on complex robotic systems, achieving up to$35 \times$improvement in runtime without any loss in performance.

This paper proposes Distributed Model Predictive Covariance Steering (DiMPCS) for multi-agent control under stochastic uncertainty. The scope of our approach is to blend covariance steering theory, distributed optimization and model predictive control (MPC) into a single framework that is safe, scalable and decentralized. Initially, we pose a problem formulation that uses the Wasserstein distance to steer the state distributions of a multi-agent system to desired targets, and probabilistic constraints to ensure safety. We then transform this problem into a finite-dimensional optimization one by utilizing a disturbance feedback policy parametrization for covariance steering and a tractable approximation of the safety constraints. To solve the latter problem, we derive a decentralized consensus-based algorithm using the Alternating Direction Method of Multipliers. This method is then extended to a receding horizon form, which yields the proposed DiMPCS algorithm. Simulation experiments on a variety of multi-robot tasks with up to hundreds of robots demonstrate the effectiveness of DiMPCS. The superior scalability and performance of the proposed method is also highlighted through a comparison against related stochastic MPC approaches. Finally, hardware results on a multi-robot platform also verify the applicability of DiMPCS on real systems.

This article considers the problem of motion coordination for a multiagent (MA) network whose goal is to track a large-scale multitarget (MT) system in a region populated by dynamic obstacles. We first characterize a density path which corresponds to the expected evolution of the macroscopic state of the MT system, which is represented by the probability density function (PDF) of a time-varying Gaussian mixture (GM). We compute this density path by using an adaptive optimal control method which accounts for the distribution of the (possibly moving) obstacles over the environment described by a time-varying obstacle map function. We show that each target of the MT system can find microscopic inputs that can collectively realize the density path while guaranteeing obstacle avoidance at all times. Subsequently, we propose a Voronoi distributed motion coordination algorithm which determines the individual microscopic control inputs of each agent of the MA network so that the latter can track the MT system while avoiding collisions with obstacles and their teammates. The proposed algorithm relies on a distributed move-to-centroid control law in which the density over the Voronoi cell of each agent is determined by the estimated macroscopic state evolution of the MT system. Finally, simulation results are presented to showcase the effectiveness of our proposed approach.

We propose a two-level coverage control framework for a multiagent network whose members have to deploy over a given region in accordance with a (possibly time-varying) coverage density function. Our approach is based on a two-level description of the multiagent network. The first level corresponds to the probability density function of the agents' locations over a given region, in which the multiagent network is treated as one unit (macroscopic description), whereas in the second level, the network is described in terms of the collection of all individual positions of its agents (microscopic description). The goal of the multiagent network is to attain a spatial distribution that (approximately) matches the reference coverage density function (high-level coverage control problem) through local interactions of the agents of the network at the individual level (low-level coverage control problem). We address the high-level control problem by associating it with an interpolation problem in the class of Gaussian mixtures. Furthermore, we address the low-level control problem by utilizing a variation of Lloyd's algorithm with a time-varying coverage density function, which is updated at each step based on the distribution of the agents' locations. Because the high-level and the low-level coverage control problems are inherently coupled to each other, we propose an iterative scheme that combines their solutions in order to address the deployment problem in a holistic way. Finally, a set of simulation results is provided to show the effectiveness of the proposed approach.

We address a planar multiagent pursuit-evasion game with a terminal constraint (reach-avoid game). Specifically, we consider the problem of steering a single evader to a target location, while avoiding capture by multiple pursuers. We propose a feasible control strategy for the evader, against a group of pursuers that adopts a semi-cooperative strategy. First, we characterize a partition of the game's state-space, that allows us to determine the existence of a solution to the game based on the initial conditions of the players. Next, based on the time-derivative of an appropriately defined risk metric, we develop a nonlinear state feedback strategy for the evader which provides a feasible solution to the game. This control strategy involves switching between different control laws in different parts of the state-space. We demonstrate the efficacy of our proposed feedback control in terms of the evader's performance, through numerical simulations. We also show that for the special case of the reach-avoid game with only one pursuer, the proposed control law is successful in guiding the evader to the target location from almost all initial conditions, and ensures that the evader will remain uncaptured.