The ability of Language Models (LMs) to understand natural language makes them a powerful tool for parsing human instructions into task plans for autonomous robots. Unlike traditional planning methods that rely on domain-specific knowledge and handcrafted rules, LMs generalize from diverse data and adapt to various tasks with minimal tuning, acting as a compressed knowledge base. However, LMs in their standard form face challenges with long-horizon tasks, particularly in partially observable multi-agent settings. We propose an LM-based Long-Horizon Planner for Multi-Agent Robotics (LLaMAR), a cognitive architecture for planning that achieves state-of-the-art results in long-horizon tasks within partially observable environments. LLaMAR employs a plan-act-correct-verify framework, allowing self-correction from action execution feedback without relying on oracles or simulators. Additionally, we present MAP-THOR, a comprehensive test suite encompassing household tasks of varying complexity within the AI2-THOR environment. Experiments show that LLaMAR achieves a 30\% higher success rate than other state-of-the-art LM-based multi-agent planners in MAP-THOR and Search \& Rescue tasks. Code can be found at [https://github.com/nsidn98/LLaMAR](https://github.com/nsidn98/LLaMAR)

We consider the problem of multi-agent navigation and collision avoidance when observations are limited to the local neighborhood of each agent. We propose InforMARL, a novel architecture for multi-agent reinforcement learning (MARL) which uses local information intelligently to compute paths for all the agents in a decentralized manner. Specifically, InforMARL aggregates information about the local neighborhood of agents for both the actor and the critic using a graph neural network and can be used in conjunction with any standard MARL algorithm. We show that (1) in training, InforMARL has better sample efficiency and performance than baseline approaches, despite using less information, and (2) in testing, it scales well to environments with arbitrary numbers of agents and obstacles. We illustrate these results using four task environments, including one with predetermined goals for each agent, and one in which the agents collectively try to cover all goals.

Integer programs provide a powerful abstraction for representing a wide range of real-world scheduling problems. Despite their ability to model general scheduling problems, solving large-scale integer programs (IP) remains a computational challenge in practice. The incorporation of more complex objectives such as robustness to disruptions further exacerbates the computational challenge. We present NICE (Neural network IP Coefficient Extraction), a novel technique that combines reinforcement learning and integer programming to tackle the problem of robust scheduling. More specifically, NICE uses reinforcement learning to approximately represent complex objectives in an integer programming formulation. We use NICE to determine assignments of pilots to a flight crew schedule so as to reduce the impact of disruptions. We compare NICE with (1) a baseline integer programming formulation that produces a feasible crew schedule, and (2) a robust integer programming formulation that explicitly tries to minimize the impact of disruptions. Our experiments show that, across a variety of scenarios, NICE produces schedules resulting in 33% to 48% fewer disruptions than the baseline formulation. Moreover, in more severely constrained scheduling scenarios in which the robust integer program fails to produce a schedule within 90 minutes, NICE is able to build robust schedules in less than 2 seconds on average.

Queuing networks have been widely-used to model congestion in transportation systems. Due to their interconnected nature, delays in a queuing network can propagate as customers traverse through the network; similarly, downstream resources can be underutilized due to poor control policies. This paper considers the regulation of arrivals into a queuing network in order to maintain a desired level of occupancy (queue length) in the system. The dynamics of the queuing network is represented by a fluid-flow model, which is then used to develop a robust controller for tracking the desired queue length. The controller is based on a sliding mode control approach, with predictor-based feedback to account for propagation delays. For a single queue, we determine sufficient conditions for tracking the queue length, and bounds on the tracking error. We also present an analysis of the tracking performance for queues in tandem. We demonstrate our approach for the example of airport surface congestion control. The proposed robust control framework is based on a queuing network model of the airport, and is used to tactically manage aircraft departures in order to reduce congestion on the airport tarmac.

This paper studies the problem of allocating tasks from different customers to vehicles in mobility platforms, which are used for applications like food and package delivery, ridesharing, and mobile sensing. A mobility platform should allocate tasks to vehicles and schedule them in order to optimize both throughput and fairness across customers. However, existing approaches to scheduling tasks in mobility platforms ignore fairness.

As the demand for Unmanned Aircraft Systems (UAS) operations increases, UAS Traffic Flow Management (UTFM) initiatives are needed to mitigate congestion, and to ensure safety and efficiency. Congestion mitigation can be achieved by assigning airborne delays (through speed changes or path stretches) or ground delays (holds relative to the desired takeoff times) to aircraft. While the assignment of such delays may increase system efficiency, individual aircraft operators may be unfairly impacted. Dynamic traffic demand, variability in aircraft operator preferences, and differences in the market share of operators complicate the issue of fairness in UTFM. Our work considers the fairness of delay assignment in the context of UTFM. To this end, we formulate the UTFM problem with fairness and show through computational experiments that significant improvements in fairness can be attained at little cost to system efficiency. We demonstrate that when operators are not aligned in how they perceive or value fairness, there is a decrease in the overall fairness of the solution. We find that fairness decreases as the air-ground delay cost ratio increases and that it improves when the operator with dominant market share has a weak preference for the fairness of its allocated delays. Finally, we implemented UTFM in a rolling-horizon setting with dynamic traffic demand, and find that efficiency is adversely impacted. However, the impact on fairness is varied and depends on the metric used.

As airport arrival capacities increasingly constrain the air transportation system, there is a need for mechanisms by which airlines can exchange landing slots amongst each other. We analyze two such mechanisms, scaled airline preferences and two-for-two trades, from a game-theoretic perspective. This paper investigates the extent to which strategic behavior on part of the airlines can impact the performance of each mechanism. In addition to increasing system efficiency, the reallocation mechanisms should exhibit desirable fairness and incentive properties, notions that we formally investigate in this paper. We show that neither mechanism has good incentive properties, and we develop simple, non-truthful strategies that airlines can use. Our empirical results show that for the scaled airline preferences mechanism, the best performing strategy depends greatly on the extent to which fairness is enforced. For the two-for-two trades mechanism, a simple threshold strategy can yield significant cost savings relative to the best-response strategy, and system efficiency increases when all airlines use the threshold strategy in equilibrium.

The runway configuration is the set of the runways at an airport that are used for arrivals and departures at any time. While many factors, including weather, expected demand, environmental considerations, and coordination of flows with neighboring airports, influence the choice of runway configuration, the actual selection decision is made by air traffic controllers in the airport tower. As a result, the capacity of an airport at any time is dependent on the behavior of human decision makers. This paper develops a statistical model to characterize the configuration selection decision process using empirical observations. The proposed approach, based on the discrete-choice modeling framework, identifies the influence of various factors in terms of the utility function of the decision maker. The parameters of the utility functions are estimated through likelihood maximization. Correlations between different alternatives are captured using a multinomial “nested logit” model. A key novelty of this study is the quantitative assessment of the effect of inertia, or the resistance to configuration changes, on the configuration selection process. The developed models are used to predict the runway configuration 3 h ahead of time, given operating conditions such as wind, visibility, and demand. Case studies based on data from Newark (EWR) and LaGuardia (LGA) airports show that the proposed model predicts runway configuration choices significantly better than a baseline model that only considers the historical frequencies of occurrence of different configurations.

Airport surface congestion control has the potential to mitigate the increase in taxi times and fuel burn at major airports. One possible class of congestion control strategies predicts the departure throughput, and recommends a rate at which to release aircraft pushbacks from the gate. This paper describes the field-testing of these types of strategies at Boston Logan International Airport, focusing on the communication of the suggested rate to the air traffic controller, and additional support for its implementation. Two Android tablet computers were used for the field-tests; one to input the data and the other to display the recommended rate to the air traffic controllers. Two potential decision-support displays were tested: a rate control display that only presented a color-coded suggested pushback rate and a volume control display that provided additional support to the controllers on the number of aircraft that had called-ready and had been released. A survey of controllers showed that they had found the decision-support tool easy to use, especially the additional functionality that is provided by the aircraft volume control display. The field tests were also found to yield significant operational benefits showing that such a congestion control strategy could be effective in practice.

Capacity constraints imposed by current air traffic management technologies and protocols could severely limit the performance of the Next Generation Air Transportation System (NextGen). A fundamental design decision in the development of this system is the level of decentralization that balances system safety and efficiency. A new surveillance technology called automatic dependent surveillance-broadcast (ADS-B) can be potentially used to shift air traffic control to a more distributed architecture; however, channel variations and interference with existing secondary radar replies can affect ADS-B systems. This paper presents a framework for managing arrivals at an airport by using a hybrid centralized/distributed algorithm for communication and control. The algorithm combines the centralized control that is used in congested regions with the distributed control that is used in lower traffic density regions. The hybrid algorithm is evaluated through realistic simulations of operations around a major airport. The proposed strategy is shown to significantly improve air traffic control performance under various operating conditions by adapting to the underlying communication, navigation, and surveillance systems. The performance of the proposed strategy is found to be comparable to fully centralized strategies, despite requiring significantly less ground infrastructure.

This paper proposes dynamic-programming-based algorithms for controlling the departure process at congested airports. These algorithms, which are called Pushback Rate Control protocols, predict the departure throughput of the airport and recommend a rate at which to release aircraft from their gates in order to control congestion. This paper describes the design and field testing of a variant of Pushback Rate Control at Boston airport in 2011 and the development of a decision support tool for its implementation. The analysis of data from the field trials shows that during eight 4-h test periods, fuel use was reduced by an estimated 9 U.S. tons (2650 U.S. gallons), and taxi-out times were reduced by an average of 5.3 min for the 144 flights that were held at the gate.

The terminal area surrounding an airport is an important component of the air transportation system, and efficient terminal-area schedules are essential for accommodating the projected increase in air traffic demand. Aircraft arrival schedules are subject to a variety of operational constraints, such as minimum separation for safety, required arrival time-windows, limited deviation from a first-come first-served sequence, and precedence constraints. There is also a range of objectives associated with multiple stakeholders that could be optimized in these schedules; the associated tradeoffs are evaluated in this paper. A dynamic programming algorithm for determining the minimum cost arrival schedule, given aircraft-dependent delay costs, is presented. The proposed approach makes it possible to determine various tradeoffs in terminal-area operations. A comparison of maximum throughput and minimum average delay schedules shows that the benefit from maximizing throughput could be at the expense of an increase in average delay, and that minimizing delay is the more advantageous of the two objectives in most cases. A comprehensive analysis of the tradeoffs between throughput and fuel costs and throughput and operating costs is conducted, accounting for both the cost of delay (as reported by the airlines) and the cost of speeding up when possible (from models of aircraft performance).