PaperlayerView exact revision on arXiv

Source-linked AI summary

Robust Multi-Agent Pursuit with Preemptive Communication via Reachability-Based Control Barrier Functions

Madison Book, Justin Bradley

arXiv:2610.22198v1eess.SYcs.MAcs.ROcs.SY

TL;DR

Safe multi-agent coordination often assumes continuous state sharing, which resource-constrained deployments may not provide. The paper combines reachability-based safety control with preemptive, risk-based communication; simulations report fewer messages while maintaining safety and capture performance.

  • Problem

    Multi-agent safety methods commonly assume continuous state information, while communication constraints and stale information complicate safe control.

    Source: §I · p. 1

  • Method

    Agents use reachable sets and CBF control for safety under intermittent information, and transmit preemptively based on uncertainty and physical risk, with a heartbeat bounding staleness.

    Source: §I · p. 1, §IV · p. 4

  • Results

    82.2% lower total message volume than baseline communication strategies, with zero inter-agent collisions.

    Source: §VI · p. 6, 5) Summary · p. 6

  • Takeaways & Limitations

    The reported results indicate that safety-aware communication can reduce message volume while maintaining safety and capture-time performance.

    Source: 5) Summary · p. 6

    Inter-agent safety requires the maximum inter-communication interval T to remain finitely bounded.

    Source: §IV · p. 4

Abstract

from arXiv · show

Safe multi-agent coordination typically assumes what real-world, deployed, resource-constrained robots cannot provide: continuous, perfect state sharing. Robust control strategies must be complemented with robust communication strategies to deliver resilient deployments. In this work, we propose a unified control/communication strategy that provides safe control to guarantee inter-agent safety under intermittent communication and demonstrate its effectiveness in a pursuit-evasion scenario. Each agent computes the reachable set of neighboring agents based on the last state information received, then solves for a safe control input based on control barrier functions to ensure agents avoid collision. We prove that inter-agent collision avoidance is guaranteed provided communication staleness remains bounded. As the reachable set inflates between state transmissions, each agent reasons about when to preemptively transmit its state information to neighboring agents based on novel metrics characterizing uncertainty risk and kinematic risk. We implement and simulate a variety of pursuit-evasion scenarios and show that preemptive communication reduces messages by 77.5% compared to periodic and event-triggered baselines with zero collisions.

I. INTRODUCTION

The paper combines reachability-based safety control with communication triggered by neighbors’ safety needs, addressing the reliance on continuous state sharing. In simulations, the unified strategy reduced message volume while maintaining safety and performance.

I. INTRODUCTION

  • Safety guarantees based on CBFs typically assume continuous state information, while bandwidth, power, or latency constraints can make that assumption fail.Prior discrete-event communication work treated communication and control separately, while stale information causes uncertainty to grow over time.Source: §I · p. 1
  • Agents transmit preemptively when neighbors need updated state information to remain safe, rather than using self-state thresholds.The approach combines reachability analysis, which over-approximates possible states since the last communication, with CBF-based safe control.Source: §I · p. 1
  • 77.5% fewer messages than comparable strategies, with the unified strategy maintaining safety and performance.The strategy is designed to account for onboard communication and computation limitations.Source: §I · pp. 1–2

II. PROBLEM FORMULATION

  • The model considers pursuers and evaders in R3, with bounded pursuer speed and control input and bounded evader acceleration.Pursuers retain the last transmitted and received neighbor states, with staleness measured from their transmission times.Source: §II · p. 2

A. Assumptions

The formulation assumes known pursuer-evader assignments and stale communicated neighbor states, with communication managed through pairwise transmissions and a heartbeat bound. Reachability sets over-approximate neighbor positions as information becomes stale.

A. Assumptions

  • Pursuers know their own and assigned evaders’ states, use fixed known assignments, and have stale communicated knowledge of other pursuers’ states.Assignments are computed using the Hungarian algorithm [14].Source: A. Assumptions · p. 2
  • Each pursuer dynamically sends state to a neighbor based on physical risk and uncertainty, with a heartbeat bounding the maximum time between transmissions.The network is idealized as directed and pairwise.Source: A. Assumptions · p. 2

1) Pursuer Objectives:

A. Reachability

  • With stale neighbor information, bounded unknown control inputs can move the true position from its predicted position; the spherical reachable set over-approximates that position.Its radius grows quadratically with staleness and resets on receipt of new state information; the stated bound places the true position inside the set.Source: A. Reachability · p. 2

1) Inter-Pursuer Safety:

The controller imposes hard pairwise pursuer safety and a soft sensing-range objective for each pursuer’s assigned evader. The soft constraint is subordinate to inter-pursuer safety.

1) Inter-Pursuer Safety:

  • The hard pairwise safety constraint keeps pursuer i at least dsafe outside pursuer j’s spherical reachable set.The barrier uses j’s reachable set and predicted position because j’s current position is unknown.Source: 1) Inter-Pursuer Safety · p. 2

2) Sensing Range Maintenance:

  • Sensing-range maintenance for an assigned evader is enforced softly through a slack variable and penalty in the QP.The soft constraint preserves the priority of inter-pursuer safety over sensing-range maintenance.Source: 2) Sensing Range Maintenance · pp. 2–3

3) High Order Control Barrier Functions:

The controller uses high-order control barrier function conditions for the relative-degree-two safety barriers. Enforcing the hard inter-pursuer constraint guarantees forward invariance of the inter-pursuer safety set.

3) High Order Control Barrier Functions:

  • The relative-degree-two barriers use high-order control barrier function conditions with positive gains a0, a1, a2, and a3.The derived controller gains are k0 = a0a1, k1 = a0 + a1, k2 = a2a3, and k3 = a2 + a3, with k1^2 > 4k0 and k3^2 > 4k2.Source: 3) High Order Control Barrier Functions · p. 3
  • Enforcing the hard inter-pursuer barrier condition guarantees forward invariance of the inter-pursuer safety set.The pursuer-evader sensing-range constraint is soft and is not required to satisfy safety.Source: 3) High Order Control Barrier Functions · p. 3

4) QP Controller:

The QP selects an optimal control input subject to hard inter-pursuer safety constraints. The proof establishes collision avoidance across communication intervals, provided the barrier conditions hold at initialization and each communication event.

4) QP Controller:

  • The QP chooses the optimal control input while the nominal input points toward the assigned evader.The hard safety constraints are enforced through the admissible control inputs; the proof carries their guarantee across communication events by induction.Source: 4) QP Controller · p. 3
  • Theorem 1 guarantees pursuer separation of at least dsafe for all time when its initial and communication-event barrier conditions hold.The theorem also requires ˙ψij + a1ψij ≥ 0 for all time, with a0 and a1 positive.Source: 4) QP Controller · p. 3
  • At each communication event, the neighbor’s prediction error resets to zero, restoring the barrier conditions needed to continue the safety guarantee.The proof applies over each communication interval and then uses continuity and induction to extend separation to all time.Source: 4) QP Controller · p. 3

IV. PREEMPTIVE COMMUNICATION

The preemptive strategy transmits state when risk indicates a neighbor may lack information to operate safely or agents face collision risk. Its risk score combines uncertainty from stale information with physical kinematic risk, while a maximum interval bounds staleness.

IV. PREEMPTIVE COMMUNICATION

  • Preemptive transmission is intended to preserve QP feasibility while minimizing unnecessary communication.QP feasibility is threatened as communication staleness grows and reachable sets inflate, reducing control authority.Source: §IV · p. 3
  • Agents transmit when estimated risk exceeds a threshold, either because a neighbor lacks enough information or because their motion presents collision risk.The risk score combines uncertainty risk and kinematic risk; the latter accounts for proximity, velocity approach, and control-input approach.Source: §IV · pp. 3–4
  • The risk threshold P trades communication frequency against uncertainty and physical risk, while maximum interval T bounds staleness and supports detection of communication failures.Inter-agent safety requires T to be finitely bounded; the strategy assumes homogeneous actuation constraints.Source: §IV · p. 4
  • Uncertainty risk measures a pursuer’s perception of a neighbor’s uncertainty about its state as staleness inflates the neighbor’s reachable set.The risk rises as that uncertainty threatens QP feasibility.Source: §IV · p. 4

V. RESULTS

The evaluation simulates cooperative pursuers pursuing adversarial evaders with intermittent neighbor-state access and compares preemptive communication with baseline mechanisms. The reported representative case illustrates risk-dependent communication and pursuit of assigned evaders.

V. RESULTS

  • Pursuers maintain predicted neighbor states from intermittent communications, while assignments to evaders are computed using the Hungarian algorithm.Agents are randomly initialized subject to the specified safe distance.Source: §V · p. 4
  • In the 3-pursuer, 2-evader case, pursuer 1 communicates more frequently with pursuer 0 than pursuer 2 because they are closer and at higher collision risk.The plotted metrics follow pursuer 1 and include distance, velocity, total risk, kinematic risk, and uncertainty risk.Source: §V · p. 5, Figure 3 · p. 5
  • The pursuer-evader distance plot shows pursuers closing on and capturing their assigned evaders.The scenario uses multiple cooperative pursuers and adversarial evaders, with intermittent access to neighboring pursuers’ states.Source: §V · pp. 4–5
  • The simulation compares no communication, full communication, periodic communication at 1 and 5 Hz, and event-triggered communication.The supplied passages describe the comparison setup but do not report comparative aggregate outcomes.Source: §V · p. 5

A. Discussion of Results

Across three pursuit-evasion scenarios, the proposed strategy achieved collision-free operation and competitive task performance with substantially lower communication costs than all baselines.

A. Discussion of Results

  • Across 8P8E, 12P6E, and 16P4E scenarios over 100 trials each, the strategy achieved collision-free operation and competitive task performance at a fraction of baseline communication costs.The comparison measured total messages, messages per agent per second, inter-pursuer collisions, and total capture time.Source: A. Discussion of Results · p. 5

1) Communication Efficiency:

The preemptive strategy reduces communication while maintaining safety and comparable capture performance. Communicating strategies preserve collision avoidance, whereas agents that do not communicate experience unavoidable collisions.

1) Communication Efficiency:

  • 29.03× fewer messages/agent/second and 96.5% lower overall message volume than full communication, with comparable safety and capture performance.The preemptive mechanism had the lowest communication overhead among strategies enforcing safety across every tested scenario.Source: 1) Communication Efficiency · p. 5
  • Against periodic baselines, total message volume fell 64.1% versus 1 Hz and 90.77% versus 5 Hz; against event-triggered communication, it fell 79.43%.Messages/agent/s were respectively 3.26×, 10.96×, and 4.78× lower than those baselines.Source: 1) Communication Efficiency · p. 5
  • The preemptive strategy reduces total message volume by 82.2% across all scenarios and baseline communication strategies, and by 77.5% compared with periodic and event-triggered mechanisms.Source: 1) Communication Efficiency · p. 5

2) Collision Avoidance:

  • Inter-agent safety remained the primary constraint and was maintained by communicating strategies across all scenarios, including under substantially reduced information availability.When agents did not communicate, collisions were unavoidable and increased with agent count.Source: 2) Collision Avoidance · pp. 5–6

3) Total Capture Time:

The strategy preserved capture performance while reducing communication, with a risk-governed communication rate that scales more favorably as agent count grows. The conclusion reports zero collisions and identifies realistic reachable sets and UAV deployment as future work.

3) Total Capture Time:

  • Capture times for the preemptive strategy and communicating baselines were within a fraction of a second in all cases.The paper states that the 82.2% message-volume reduction incurred no significant penalty in total capture or task completion time.Source: 3) Total Capture Time · p. 6

4) Scalability:

  • The preemptive mechanism’s communication rate is governed by situational risk, yielding more favorable scaling as the number of agents grows.Full and periodic strategies scale with agent pairs and communication interval; event-triggered communication scales with significant self-state changes.Source: 4) Scalability · p. 6

5) Summary:

  • Across scenarios and baseline communication strategies, the preemptive mechanism reduced total message volume by 82.2% while matching baseline safety and capture-time performance.The paper presents this as evidence that safety-aware communication can reduce communication while maintaining provable safety guarantees.Source: 5) Summary · p. 6

VI. CONCLUSION

  • 82.2% lower total message volume than baseline communication strategies while maintaining zero inter-agent collisions.The reachability-based HOCBF-QP controller enforces inter-agent safety under intermittent state information.Source: §VI · p. 6
  • More realistic reachable-set formulations and deployment on SWaP-constrained UAV platforms remain future work for validating real-world effectiveness.The paper identifies both as future directions.Source: §VI · p. 6

Related papers

Loading 2610.22198v1…