Source-linked AI summary
A Critical Review of Communications in Multi-Robot Systems
Jennifer Gielis, Ajay Shankar, Amanda Prorok
TL;DR
Multi-robot systems depend on communication, yet robotics and networking are often optimized separately and existing communication strategies rely on idealized assumptions. This review surveys applications and network technologies, examines data-driven communication approaches, and identifies co-optimization and sim-to-real challenges alongside open problems for future work.
Problem
Robotics and networking are rarely co-designed, while communication strategies often assume ideal operations and leave transferable handling of loss, delay, and corruption unclear.
Method
The review synthesizes communication technologies, network architectures, multi-robot applications, joint optimization approaches, and learning-based communication strategies.
Results
The review identifies deficiencies in the evolution of robotics and networks and presents machine learning as an avenue for developing data-driven communication strategies.
Takeaways & Limitations
A more synergistic approach should consider perception, control, communications, and their confounding factors together in multi-robot system optimization.
Takeaways & Limitations
Communication reality gaps involving dropouts, asynchronous or out-of-order reception, and unreliable decentralized mesh topologies are difficult to capture in typically synchronous training and remain sparsely studied at large scale.
Abstract
from arXiv · showhide
Purpose of Review. This review summarizes the broad roles that communication formats and technologies have played in enabling multi-robot systems. We approach this field from two perspectives: of robotic applications that need communication capabilities in order to accomplish tasks, and of networking technologies that have enabled newer and more advanced multi-robot systems. Recent Findings. Through this review, we identify a dearth of work that holistically tackles the problem of co-design and co-optimization of robots and the networks they employ. We also highlight the role that data-driven and machine learning approaches play in evolving communication pipelines for multi-robot systems. In particular, we refer to recent work that diverges from hand-designed communication patterns, and also discuss the "sim-to-real" gap in this context. Summary. We present a critical view of the way robotic algorithms and their networking systems have evolved, and make the case for a more synergistic approach. Finally, we also identify four broad Open Problems for research and development, while offering a data-driven perspective for solving some of them.
1. Introduction
Multi-robot applications use communication to coordinate distributed work, share information, and pursue common goals, but current systems often adapt to network constraints rather than co-designing robots and networks together. This review surveys communication technologies and applications while motivating simultaneous architectural co-optimization and data-driven communication.
- Connected robots enable work distribution, spatial coverage, specialization, and coordination across applications including logistics, transport, manufacturing, and agriculture.
- Communication exchanges robot state and control information for joint goals, action deconfliction, and distributed computing.
- Early systems used local interactions, near-field transmission, and stigmergy to produce decentralized collective behavior.
- Dense workspaces, signal fading, and real-time requirements make robot network conditions hostile to radio performance and reduce reliable communication range.
- Most robot applications work around available network technologies and optimize within their constraints instead of jointly optimizing networking and autonomy.
- The review surveys network architectures, technologies, and applications while examining data-driven approaches that model inter-robot communication beyond hand-crafted selections.
2. Factors Influencing Robot Network Design
Robot network design depends on application, robot, algorithm, and environmental requirements that determine what information must be exchanged, when, and under which delivery conditions. These factors impose distinct spatial, temporal, computational, and connectivity constraints on communication.
- Network design therefore must match application, robot, algorithm, and environment factors rather than selecting communication in isolation.
- The application: Application requirements determine the shared information and interaction pattern, including long-range networking for robots separated by approximately 1000x body lengths.
- The robot: Robot hardware constrains the frequency and format of exchanged information, with quadrotor stabilization requiring updates at several hundred Hertz and extrinsic pose estimates around 100 Hz.
- The robot: Unreliable updates risk tightly coordinated tasks such as outdoor flocking and formation control, while GPS-denied group behavior requires dependable inter-robot communication.
- The algorithm: Algorithms determine whether information is global or local and whether it arrives synchronously or asynchronously, including centralized collection of assignment costs for allocation.
- The environment: Environmental conditions determine infrastructure, topology, signal interference, communication medium, and the delivery conditions for shared information.
3. Communication Schemes
Multi-robot communication schemes face timing, contention, range, interference, and routing constraints that existing wireless standards do not fully address. The review surveys ad-hoc, local-area, cellular, and emerging access technologies in relation to these challenges.
- Communication challenges: Synchronous control imposes hard message-delay limits, while dynamic, highly connected topologies increase contention as robot teams scale.Each control loop may require many status messages, making communication technology selection and design increasingly important.
- Communication challenges: Bandwidth alone is insufficient because per-message overhead and low-latency requirements constrain communication performance, especially in ad-hoc networks.Message frequency and protocol overhead can prevent systems from approaching nominal link bandwidth.
- Communication challenges: Longer communication ranges increase exposure to fading and interference, while multi-hop routing adds processing overhead and lacks hard timing guarantees at non-trivial scales.Dynamic topologies make routing especially difficult in highly mobile or large-scale robotic scenarios.
- Communication technologies: No wireless data standard is explicitly designed for autonomous-robot information exchange, so applications commonly rely on generic technologies not optimized for robotic constraints.The review therefore compares existing standards by their strengths and weaknesses for multi-robot control.
- Communication technologies: Wi-Fi and IEEE 802.15.4 offer practical deployment advantages, but their contention-avoidance mechanisms provide unbounded latency and lower delivery rates as robot numbers increase.LoRaWAN scales even more poorly because it uses pure ALOHA, leaving these protocols unsuitable without modifications.
- Communication technologies: Even OFDMA-based 5G and 802.11ax may reach CSMA/CA limits in extremely dense ad-hoc robot networks, leaving decentralized contention unresolved.NOMA may expand simultaneous radio resources, but coordination and message loss remain open problems without infrastructure.
4. Communication-Aware Algorithms
Communication-aware multi-robot algorithms incorporate imperfect links into planning, schedule communication as part of the task, or jointly optimize paths and communication. The reviewed work spans connectivity-aware control, constrained environments, and explicit co-optimization strategies.
- Communication-aware planning: Many multi-robot control schemes still omit communication dependencies, a challenge that becomes more pronounced when robots must deconflict and replan in constrained spaces.Recent work considers communication asynchronicity, but the broader challenge remains unresolved.
- Planning and communication strategies: Communication constraints can be handled by reducing transmitted data, modeling channels in motion planning, scheduling communication, or jointly optimizing paths and communication.The review organizes this literature into three broad styles: communication-constrained planning, communication as a sub-task, and joint optimization.
- Communication-aware planning: Formation and coverage control methods explicitly factor communication constraints through latency-aware stability analysis, limited communication ranges, and connectivity requirements.Related applications include cooperative target localization and operation under 3G/4G mobile-network constraints.
- Plan-aware communications: Constrained environments motivate architectures that seek communication opportunities along planned paths, including underwater networks with severe bandwidth and interference limits.Scheduling methods model the value of communicating at particular times and support decisions about whether and what to communicate.
- Joint optimization: Joint optimization approaches include sequentially alternating path and communication optimization and cyber-physical-system models coupling communication and kinematic controllers.CPS formulations can adapt communication capacity while maintaining coupling with the physical controller.
5. Leveraging Machine Learning for Communication
The review presents machine learning as an emerging alternative to handcrafted communication design, addressing decentralized decision-making under partial observability and dynamic network conditions. It discusses learning-based communication, graph-based representations, and the need to guard against adversarial behaviors.
- Motivation: Handcrafted decentralized policies remain difficult to synthesize and often leave unanswered what, when, and to whom robots should communicate.Existing strategies also commonly rely on idealized assumptions about message loss, delay, and corruption.
- Networking Applications: Learning methods offer alternatives for dynamic routing, channel modeling, and resource allocation where manually specified heuristics may rely on incorrect environmental assumptions.Unsupervised learning can estimate link quality to optimize transmission power.
- Learning Communication Behaviors: Robot-to-robot communication learning is nascent because each robot’s local decision system lacks other agents’ unobservable states that affect future values.The review frames learning as a way to balance centralized full-information quality with decentralized partial-information efficiency.
- Learning Communication Behaviors: Permutation equivariance helps decentralize centralized policies by ensuring actions rearrange consistently when agents change order.Graph Neural Networks provide this property by design.
- Graph-Based Communication: Graph Neural Networks model robots as nodes, communication links as edges, and robot states as graph signals while compressing information exchanged through the network.This representation gives agents access to relevant information from the broader network.
- Multi-Agent Reinforcement Learning: Multi-agent reinforcement learning formulates continuous communication as a Dec-POMDP, including learned message relevance and targeted communication strategies.The review cautions that learned communication can also produce manipulative, non-cooperative behavior between teams.
6. Challenges and Open Problems
The review identifies four open problems: jointly designing robots and networks, expanding data-driven optimization, addressing communication-specific sim-to-real gaps, and developing technologies tailored to robotic networks. These problems reflect the need to align communication resources and robot control with real-world deployment conditions.
- Co-design: Co-design should jointly evolve networking layers and robot controllers so communication capabilities and robotic task requirements are managed together.The review describes dynamic throttling of both requirements and capabilities as relevant to safe deployment.
- Co-design: Explicitly communicating robot controllers can fail when networks under-deliver and perform sub-optimally when networks over-deliver.This creates a communication-resource allocation problem for real-world multi-robot systems.
- Data-driven optimization: Data-driven optimization could learn communication decisions and jointly optimize perception, action, and communication when sufficiently large datasets are available.Current approaches still often depend on hand-designed architectures and specific task groups.
- Sim-to-real of robot networks: Robot-network sim-to-real transfer is complicated by communication dropouts, asynchronous and out-of-order reception, and unreliable decentralized mesh topologies.The review identifies embedding this communication reality gap into data-driven approaches as an open research domain.
- New technologies/schemes: Wireless standards designed for connected robots are needed, with the IEEE 1920 working group identified as a step toward autonomous robotic-network protocols.The review associates this direction with 802.11bd rather than legacy 802.11p.
- New technologies/schemes: Future 5G and 6G updates could bring cloud and edge computing to data-intensive multi-robot collaborations.Geographic routing in FANETs is also identified as a possible response to highly dynamic routing topologies, alongside coordinated robot-control development.
7. Conclusion
The paper surveys communication technologies and their roles in multi-robot applications while critically examining how robotics systems handle network constraints. It highlights data-driven communication strategies and concludes with challenges, open problems, and learning-based directions.
- Scope and critique: The survey covers communication technologies, networked-robotics applications, and the ways state-of-the-art robot systems handle network constraints.Its critical perspective identifies deficiencies in how robotics and networks have evolved.
- Data-driven directions: The paper discusses machine learning approaches for developing data-driven communication strategies in multi-robot systems.It also presents challenges and open problems together with an outlook for learning-based approaches.
Declarations
The declarations report no conflicts of interest and no studies involving human or animal subjects.
- Declarations: The authors declare no conflicts of interest and report no studies involving human or animal subjects.