Design and Implementation of a Distributed Executive

Abstract

The deployment of autonomous robots has the potential to revolutionize high-risk missions, from rescue operations and outer-space exploration to the maintenance of underwater infrastructure. In many of these scenarios, such as the routine maintenance of a distant space station, collaboration between multiple robots becomes essential. However, the vastness of space or the depths of the oceans often impose communication constraints, making realtime coordination challenging. While centralized control of these missions is traditional and straightforward to implement, it often becomes impractical in these contexts because of communication delays and uncertainties. Given these challenges, a distributed approach is not just preferred but necessary, ensuring robots can operate independently when under limited communication conditions. Traditional strategies to coordinate multiple robots’ schedules when communication is unreliable have been conservative, as they tend to favor prefixing the time for their actions, creating rigid and non-robust schedules. Such schedules can allocate excessive time to tasks as a safety measure, leaving potential resources underutilized. This over-caution not only results in inefficient execution but can also prevent the executive from identifying viable schedules for missions, even when they exist with a more flexible approach. The lack of adaptability, especially in the face of unexpected challenges, undermines the executive’s robustness. To address these shortcomings, our aim is to craft a flexible and robust distributed executive adept at planning, scheduling, and executing multi-agent scenarios. We build upon the Kirk executive, a creation of the MERS group at CSAIL, enabling it to proficiently manage multi-agent scenarios without a guarantee of perfect communication during execution. Central to our methodology is the principle of temporal decoupling which allows agents to decouple any inter-dependencies in their schedule and operate independently. We integrate the state of the art algorithm in temporal decoupling, which decouples as necessary, leaving room for communication when it is available. This integration not only enhances the autonomy of the agents but also ensures they can leverage the benefits of communication when it is available, striking a balance between independence and collaborative efficiency. Building on this foundation, our work offers a practical perspective on autonomous robot coordination. By enhancing the Kirk executive with a temporal decoupling algorithm, expanding the Reactive Model-based Programming Language (RMPL) for multi-agent scenario representation, and showcasing Kirk’s improved capability in multi-agent scenarios with communication constraints, we bridge the gap between theoretical foundations and practical applications.