On-board Mission Replanning for Adaptive Cooperative Multi-Robot Systems (2506.06094v1)

Abstract: Cooperative autonomous robotic systems have significant potential for executing complex multi-task missions across space, air, ground, and maritime domains. But they commonly operate in remote, dynamic and hazardous environments, requiring rapid in-mission adaptation without reliance on fragile or slow communication links to centralised compute. Fast, on-board replanning algorithms are therefore needed to enhance resilience. Reinforcement Learning shows strong promise for efficiently solving mission planning tasks when formulated as Travelling Salesperson Problems (TSPs), but existing methods: 1) are unsuitable for replanning, where agents do not start at a single location; 2) do not allow cooperation between agents; 3) are unable to model tasks with variable durations; or 4) lack practical considerations for on-board deployment. Here we define the Cooperative Mission Replanning Problem as a novel variant of multiple TSP with adaptations to overcome these issues, and develop a new encoder/decoder-based model using Graph Attention Networks and Attention Models to solve it effectively and efficiently. Using a simple example of cooperative drones, we show our replanner consistently (90% of the time) maintains performance within 10% of the state-of-the-art LKH3 heuristic solver, whilst running 85-370 times faster on a Raspberry Pi. This work paves the way for increased resilience in autonomous multi-agent systems.

• The paper introduces the Cooperative Mission Replanning Problem (CMRP), an adaptation of the mTSP addressing multiple start points, variable task durations, and cooperative tasking.
• It develops the Graph Attention Replanner (GATR), an encoder-decoder model that leverages reinforcement learning with attention mechanisms for real-time re-planning.
• GATR achieves solutions within 10% of state-of-the-art heuristics while operating 85 to 370 times faster on edge devices, ensuring mission continuity in dynamic settings.

On-board Mission Replanning for Adaptive Cooperative Multi-Robot Systems

In the field of multi-agent systems, the paper titled "On-board Mission Replanning for Adaptive Cooperative Multi-Robot Systems" addresses a critical gap in enhancing the resilience and adaptability of autonomous robotic systems operating in dynamic and remote environments. These systems, encompassing UAVs, UUVs, USVs, UGVs, and satellite constellations, have significant potential for efficiently executing complex missions. However, the dynamic constraints of these environments necessitate rapid on-board replanning capabilities, independent of centralized control systems.

The authors introduce the Cooperative Mission Replanning Problem (CMRP), a novel variant of the multiple Traveling Salesperson Problem (mTSP), specifically tailored for these requirements. CMRP is defined by three critical adaptations over traditional mTSP: multiple start locations for agents, tasks with variable time costs, and cooperative tasking allowing asynchronous task completion by multiple agents. These adaptations are crucial for addressing unpredictable mission changes such as task modifications or agent failures, where centralized communication might be inadequate due to bandwidth limitations or latency issues.

The paper presents a robust solution through the development of a Graph Attention Network-based model, the Graph Attention Replanner (GATR). This encoder/decoder architecture effectively integrates these adaptations using reinforcement learning frameworks with attention mechanisms. The model leverages Graph Attention and Attention Model architectures, with its efficacy demonstrated through extensive testing with aerial drones in a simulated environment.

Significantly, the GATR model achieves superior performance, with solutions consistently within 10% of the state-of-the-art LKH3 heuristic solver's output while operating at an impressive speed—85 to 370 times faster on a Raspberry Pi, demonstrating its practical utility for real-time on-board replanning. This rapid execution on edge devices is pivotal for maintaining mission continuity in the face of operational uncertainties, thereby enhancing overall system resilience.

The implications of this research extend beyond immediate applications. The GATR model's adaptive framework can generalize across varying agent and task scenarios, showcasing its versatility in different operational contexts. Furthermore, its lightweight and efficient architecture sets the groundwork for future advancements in AI, where similar reinforcement learning strategies could be employed to tackle other complex, dynamic systems.

Looking forward, this research could inspire developments in multi-objective planning within diverse environments, where trade-offs between mission time, task completion rates, and system resource management can be dynamically optimized. Moreover, incorporating probabilistic models could further enhance decision-making under uncertainty, fostering the development of consensus-based and anticipatory planning frameworks for collaborative robotic systems.

This paper contributes substantially to the strategic vision of autonomous systems capable of adaptive, on-the-fly mission replanning, reinforcing the viability of deploying intelligent agents in challenging environments without the necessity for continual operator intervention or reliance on centralized control systems.