Deploying Ten Thousand Robots: Scalable Imitation Learning for Lifelong Multi-Agent Path Finding

Lifelong Multi-Agent Path Finding (LMAPF) is a variant of MAPF where agents are continually assigned new goals, necessitating frequent re-planning to accommodate these dynamic changes. Recently, this field has embraced learning-based methods, which reactively generate single-step actions based on individual local observations. However, it is still challenging for them to match the performance of the best search-based algorithms, especially in large-scale settings. This work proposes an imitation-learning-based LMAPF solver that introduces a novel communication module and systematic single-step collision resolution and global guidance techniques. Our proposed solver, Scalable Imitation Learning for LMAPF (SILLM), inherits the fast reasoning speed of learning-based methods and the high solution quality of search-based methods with the help of modern GPUs. Across six large-scale maps with up to 10,000 agents and varying obstacle structures, SILLM surpasses the best learning- and search-based baselines, achieving average throughput improvements of 137.7% and 16.0%, respectively. Furthermore, SILLM also beats the winning solution of the 2023 League of Robot Runners, an international LMAPF competition sponsored by Amazon Robotics. Finally, we validated SILLM with 10 real robots and 100 virtual robots in a mockup warehouse environment.

Multi-Robot Motion Planning with Diffusion Models

Diffusion models have recently been successfully applied to a wide range of robotics applications for learning complex multi-modal behaviors from data. However, prior works have mostly been confined to single-robot and small-scale environments due to the high sample complexity of learning multi-robot diffusion models. In this paper, we propose a method for generating collision-free multi-robot trajectories that conform to underlying data distributions while using only single-robot data. Our algorithm, Multi-robot Multi-model planning Diffusion (MMD), does so by combining learned diffusion models with classical search-based techniques -- generating data-driven motions under collision constraints. Scaling further, we show how to compose multiple diffusion models to plan in large environments where a single diffusion model fails to generalize well. We demonstrate the effectiveness of our approach in planning for dozens of robots in a variety of simulated scenarios motivated by logistics environments. View video demonstrations in our supplementary material, and our code at: https://github.com/yoraish/mmd.

Windowed MAPF with Completeness Guarantees

Traditional multi-agent path finding (MAPF) methods try to compute entire start-goal paths which are collision free. However, computing an entire path can take too long for MAPF systems where agents need to replan fast. Methods that address this typically employ a "windowed" approach and only try to find collision free paths for a small windowed timestep horizon. This adaptation comes at the cost of incompleteness; all current windowed approaches can become stuck in deadlock or livelock. Our main contribution is to introduce our framework, WinC-MAPF, for Windowed MAPF that enables completeness. Our framework uses heuristic update insights from single-agent real-time heuristic search algorithms as well as agent independence ideas from MAPF algorithms. We also develop Single-Step CBS (SS-CBS), an instantiation of this framework using a novel modification to CBS. We show how SS-CBS, which only plans a single step and updates heuristics, can effectively solve tough scenarios where existing windowed approaches fail.

Work Smarter Not Harder: Simple Imitation Learning with CS-PIBT Outperforms Large Scale Imitation Learning for MAPF

Multi-Agent Path Finding (MAPF) is the problem of effectively finding efficient collision-free paths for a group of agents in a shared workspace. The MAPF community has largely focused on developing high-performance heuristic search methods. Recently, several works have applied various machine learning (ML) techniques to solve MAPF, usually involving sophisticated architectures, reinforcement learning techniques, and set-ups, but none using large amounts of high-quality supervised data. Our initial objective in this work was to show how simple large scale imitation learning of high-quality heuristic search methods can lead to state-of-the-art ML MAPF performance. However, we find that, at least with our model architecture, simple large scale (700k examples with hundreds of agents per example) imitation learning does \textit{not} produce impressive results. Instead, we find that by using prior work that post-processes MAPF model predictions to resolve 1-step collisions (CS-PIBT), we can train a simple ML MAPF model in minutes that dramatically outperforms existing ML MAPF policies. This has serious implications for all future ML MAPF policies (with local communication) which currently struggle to scale. In particular, this finding implies that future learnt policies should (1) always use smart 1-step collision shields (e.g. CS-PIBT), (2) always include the collision shield with greedy actions as a baseline (e.g. PIBT) and (3) motivates future models to focus on longer horizon / more complex planning as 1-step collisions can be efficiently resolved.

Benchmarking Large Neighborhood Search for Multi-Agent Path Finding

Multi-Agent Path Finding (MAPF) aims to arrange collision-free goal-reaching paths for a group of agents. Anytime MAPF solvers based on large neighborhood search (LNS) have gained prominence recently due to their flexibility and scalability. Neighborhood selection strategy is crucial to the success of MAPF-LNS and a flurry of methods have been proposed. However, several pitfalls exist and hinder a comprehensive evaluation of these new methods, which mainly include: 1) Lower than actual or incorrect baseline performance; 2) Lack of a unified evaluation setting and criterion; 3) Lack of a codebase or executable model for supervised learning methods. To overcome these challenges, we conduct a fair comparison across prominent methods on the same benchmark and hyperparameter search settings. Additionally, we propose a simple neighborhood selection strategy which marks a clear advancement in terms of runtime efficiency in large maps with large number of agents. Our benchmarking evaluation promotes new challenges for existing learning based methods and presents opportunities for future research when machine learning is integrated with MAPF-LNS. Code and data are available at https://github.com/ChristinaTan0704/mapf-lns-benchmark.

LNS2+RL: Combining Multi-agent Reinforcement Learning with Large Neighborhood Search in Multi-agent Path Finding

Multi-Agent Path Finding (MAPF) is a critical component of logistics and warehouse management, which focuses on planning collision-free paths for a team of robots in a known environment. Recent work introduced a novel MAPF approach, LNS2, which proposed to repair a quickly-obtainable set of infeasible paths via iterative re-planning, by relying on a fast, yet lower-quality, priority-based planner. At the same time, there has been a recent push for Multi-Agent Reinforcement Learning (MARL) based MAPF algorithms, which let agents learn decentralized policies that exhibit improved cooperation over such priority planning, although inevitably remaining slower. In this paper, we introduce a new MAPF algorithm, LNS2+RL, which combines the distinct yet complementary characteristics of LNS2 and MARL to effectively balance their individual limitations and get the best from both worlds. During early iterations, LNS2+RL relies on MARL for low-level re-planning, which we show eliminates collisions much more than a priority-based planner. There, our MARL-based planner allows agents to reason about past and future/predicted information to gradually learn cooperative decision-making through a finely designed curriculum learning. At later stages of planning, LNS2+RL adaptively switches to priority-based planning to quickly resolve the remaining collisions, naturally trading-off solution quality and computational efficiency. Our comprehensive experiments on challenging tasks across various team sizes, world sizes, and map structures consistently demonstrate the superior performance of LNS2+RL compared to many MAPF algorithms, including LNS2, LaCAM, and EECBS, where LNS2+RL shows significantly better performance in complex scenarios. We finally experimentally validate our algorithm in a hybrid simulation of a warehouse mockup involving a team of 100 (real-world and simulated) robots.

Unconstraining Multi-Robot Manipulation: Enabling Arbitrary Constraints in ECBS with Bounded Sub-Optimality

Multi-Robot-Arm Motion Planning (M-RAMP) is a challenging problem featuring complex single-agent planning and multi-agent coordination. Recent advancements in extending the popular Conflict-Based Search (CBS) algorithm have made large strides in solving Multi-Agent Path Finding (MAPF) problems. However, fundamental challenges remain in applying CBS to M-RAMP. A core challenge is the existing reliance of the CBS framework on conservative "complete" constraints. These constraints ensure solution guarantees but often result in slow pruning of the search space -- causing repeated expensive single-agent planning calls. Therefore, even though it is possible to leverage domain knowledge and design incomplete M-RAMP-specific CBS constraints to more efficiently prune the search, using these constraints would render the algorithm itself incomplete. This forces practitioners to choose between efficiency and completeness. In light of these challenges, we propose a novel algorithm, Generalized ECBS, aimed at removing the burden of choice between completeness and efficiency in MAPF algorithms. Our approach enables the use of arbitrary constraints in conflict-based algorithms while preserving completeness and bounding sub-optimality. This enables practitioners to capitalize on the benefits of arbitrary constraints and opens a new space for constraint design in MAPF that has not been explored. We provide a theoretical analysis of our algorithms, propose new "incomplete" constraints, and demonstrate their effectiveness through experiments in M-RAMP.

Scaling Lifelong Multi-Agent Path Finding to More Realistic Settings: Research Challenges and Opportunities

Multi-Agent Path Finding (MAPF) is the problem of moving multiple agents from starts to goals without collisions. Lifelong MAPF (LMAPF) extends MAPF by continuously assigning new goals to agents. We present our winning approach to the 2023 League of Robot Runners LMAPF competition, which leads us to several interesting research challenges and future directions. In this paper, we outline three main research challenges. The first challenge is to search for high-quality LMAPF solutions within a limited planning time (e.g., 1s per step) for a large number of agents (e.g., 10,000) or extremely high agent density (e.g., 97.7%). We present future directions such as developing more competitive rule-based and anytime MAPF algorithms and parallelizing state-of-the-art MAPF algorithms. The second challenge is to alleviate congestion and the effect of myopic behaviors in LMAPF algorithms. We present future directions, such as developing moving guidance and traffic rules to reduce congestion, incorporating future prediction and real-time search, and determining the optimal agent number. The third challenge is to bridge the gaps between the LMAPF models used in the literature and real-world applications. We present future directions, such as dealing with more realistic kinodynamic models, execution uncertainty, and evolving systems.

From Space-Time to Space-Order: Directly Planning a Temporal Planning Graph by Redefining CBS

The majority of multi-agent path finding (MAPF) methods compute collision-free space-time paths which require agents to be at a specific location at a specific discretized timestep. However, executing these space-time paths directly on robotic systems is infeasible due to real-time execution differences (e.g. delays) which can lead to collisions. To combat this, current methods translate the space-time paths into a temporal plan graph (TPG) that only requires that agents observe the order in which they navigate through locations where their paths cross. However, planning space-time paths and then post-processing them into a TPG does not reduce the required agent-to-agent coordination, which is fixed once the space-time paths are computed. To that end, we propose a novel algorithm Space-Order CBS that can directly plan a TPG and explicitly minimize coordination. Our main theoretical insight is our novel perspective on viewing a TPG as a set of space-visitation order paths where agents visit locations in relative orders (e.g. 1st vs 2nd) as opposed to specific timesteps. We redefine unique conflicts and constraints for adapting CBS for space-order planning. We experimentally validate how Space-Order CBS can return TPGs which significantly reduce coordination, thus subsequently reducing the amount of agent-agent communication and leading to more robustness to delays during execution.

ITA-ECBS: A Bounded-Suboptimal Algorithm for Combined Target-Assignment and Path-Finding Problem

Multi-Agent Path Finding (MAPF), i.e., finding collision-free paths for multiple robots, plays a critical role in many applications. Sometimes, assigning a specific target to each agent also presents a challenge. The Combined Target-Assignment and Path-Finding (TAPF) problem, a variant of MAPF, requires simultaneously assigning targets to agents and planning collision-free paths. Several algorithms, including CBM, CBS-TA, and ITA-CBS, can optimally solve the TAPF problem, with ITA-CBS being the leading method of flowtime. However, the only existing suboptimal method ECBS-TA, is derived from CBS-TA rather than ITA-CBS, and adapting the optimal ITA-CBS method to its bounded-suboptimal variant is a challenge due to the variability of target assignment solutions in different search nodes. We introduce ITA-ECBS as the first bounded-suboptimal variant of ITA-CBS. ITA-ECBS employs focal search to enhance efficiency and determines target assignments based on a new lower bound matrix. We show that ITA-ECBS outperforms the baseline method ECBS-TA in 87.42% of 54,033 test cases.