Fast-time Simulation Research Articles

Validating Flow-Based Arrival Management for En Route Airspace: Human-In-The-Loop Simulation Experiment with ESCAPE Light Simulator

The advancement of Arrival MANager (AMAN) is crucial for addressing the increasing complexity and demand of modern airspace. This study evaluates the operational feasibility and effectiveness of an innovative AMAN designed for en route airspace, the so-called En Route AMAN. The En Route AMAN functions as a controller support system, facilitating the sharing of information between en route air traffic controllers (ATCos), approach controllers (current AMAN), and airport controllers (Departure Managers) in airports with multiple runways. The En Route AMAN aims to support upstream ATCos by sequencing and spacing of incoming streams via speed control and runway assignment, thereby enhancing overall air traffic efficiency. Human-In-The-Loop simulations involving rated ATCos are performed under scenarios that replicate real-world traffic and weather conditions. These simulations focus on upstream airspace to assess the impact of En Route AMAN on delay mitigation and ATCos’ performance. Unlike previous studies that solely relied on theoretical models and fast-time simulation for operational feasibility evaluation, this approach incorporates ATCos’ real-time decision-making, situational awareness, and task management, addressing critical operationalization challenges. The results demonstrated that the En Route AMAN could reduce the average flight duration by up to 25.6 s and decrease the total number of ATCo instructions by up to 20% during peak traffic volume. These findings support that the En Route AMAN is both operationally viable and effective in mitigating arrival delays, highlighting the importance of Human-In-The-Loop for practical validation.

Scenarios of Multiple Vessels Encounter for Safety Evaluation for Automatic Collision Avoidance Algorithm using AIS Data

Abstract Interest in Maritime Autonomous Surface Ships (MASS) is growing, and efforts are underway in Japan to realize them. One of the key issues in realizing autonomous operation is automatic collision avoidance. Simulation is effective in developing and evaluating such algorithms. This paper describes the generation of a set of simulation scenarios that consider the actual sea traffic using AIS data, which is traffic flow data in real seas. We assumed that the comprehensiveness of the scenarios could be ensured by generating scenarios by combining typical encounter patterns and collecting difficult situations that are encountered in real situations. First, the characteristics of the target sea area were identified from the AIS data, then difficult traffic situations were extracted from the data, and then a set of scenarios were generated based on them. Considering the presence of manned ships in the test area, we conducted some tests of the generated scenarios not only using fast-time simulations but also using a ship-handling simulator and conducted visual checks from own ship and the target ship.

Manoeuvring in shallow and confined water with model predictive track controllers

Abstract Autonomous and computer-assisted navigation, for example through portable pilot units and data-driven decision supportive systems, is in full development. Remote-controlled vessels (manned but without a captain steering continuously on board) are used in Belgium (by Seafar in inland shipping) but also in other countries. A model predictive controller has been used to investigate environmental and operational parameters for the passage of the river Seine in Paris by different ship types. These fast-time simulations have been compared with simulations in real time executed by skippers on a ship manoeuvring simulator or measured real life tracks. A good predictability of the real situation requires accurate manoeuvring models. But what if this accuracy cannot be guaranteed? An illustration is made by comparing the tracks if, in the prescience prediction phase for the controller, wind or bank effects are neglected or doubled. The development of model predictive deep reinforcement learning control algorithms for navigation will be one of the challenges of the Data-driven Smart Shipping project that started May 2024.

Preliminary Impact Assessment of the Ad Hoc Separation Minima: A New Separation Mode

A major challenge for the Air Traffic Management system is the need to boost airspace capacity, which is near saturation in some situations. Separation minima are one of the factors related to airspace capacity, and the SESAR program promotes research into advanced separation modes. Ad Hoc separation, a novel separation mode, involves applying different pairwise separation minima in the same volume of airspace depending on a set of factors, such as encounter geometry, aircraft models, and flight level, among others. This research examines the impact of implementing this concept in different en-route scenarios. The goal is to determine whether applying this concept proves advantageous or, conversely, results in an increase in the complexity of the system without significant benefits in the key performance areas of capacity, environment, and cost-efficiency. Fast Time Simulations are conducted in RAMS software, with the concept being implemented in the LECMZMU, LECMTLU, and LECMDGU sectors of the Madrid ACC. The results reveal favorable capacity outcomes with increases of around 2% and the LECMZMU sector exhibits the most significant environmental and cost-efficiency benefits. Furthermore, implementing the Ad Hoc concept in a larger scenario could yield even greater environmental and cost-efficiency benefits.

Modeling extended L-band fiber amplifiers using neural networks trained on experimental data

Producing high performance amplifiers requires accurate numerical models. As the optimization space is large, computationally efficient models are of great value. Parameter-based models for L-band amplifiers have accuracy limited by difficulty in estimating the Giles-parameter. The use a neural network model can avoid parametrization. We exploit a rich, experimentally captured training set to achieve a high accuracy neural network model. Our approach creates independent models for gain and noise figure. We examine both core and cladding pumping methods, again with independent models for each. The neural networks outperform parameter-based models with higher accuracy (variance of error reduced by 50%) and extremely fast simulation times (400 times faster), greatly facilitating amplifier design. As an example application, we design an amplifier to optimize optical signal-to-noise ratio by exhaustive search with our fast neural network models.

Analysis of cross-border free route implementation: A case for Turkish and Ukrainian airspaces

Abstract Free route airspace (FRA) are increasingly widespread in European countries. A step further, cross-border FRA aims to implement FRA in a seamless,defragmented airspace in Europe. This study aims to bring a different perspective by extending the scope of cross-border FRA applications beyond socioeconomic status (SES) countries. In the study, analyses were made using fast-time simulation outputs for one of the busiest days of Turkish airspace in 2022. The simulation models of four cases were created and compared in such a way that airspaces of Türkiye, South East Europe free route airspace (SEEFRA) countries and Ukraine combined under different border operations above FL285. Flight trajectories were assessed in terms of flight distance, fuel economy and environmental impact within the aforementioned airspaces. The participation of Turkish airspace in the cross-border region and opening of Ukrainian airspace are examined. Concluded, among all, cross-border FRA implemented through all aforementioned regions (Case 4) may constitute the best benefit for flight trajectories and related variables.

A Hybrid Electrochemical Multi-Particle Model for Lithium-Ion Batteries

Over the years, physics-based models have proven to be effective tools in understanding the behavior of lithium-ion batteries, which is favorable for enhancing their design and performance. Among the various physics-based models, the Doyle-Fuller-Newman (DFN) model [1] has emerged as the most widely used model due to its ability to simulate battery behavior accurately. However, it has been observed that the DFN model performs poorly in the relaxation region after discharge, which might result in inaccurate battery simulations. The Multiple-Particle DFN (MP-DFN) model [2] has been proposed to address this issue. The MP-DFN model uses multiple electrode particle sizes to allow internal concentration heterogeneities, accurately capturing the slow diffusion processes. However, the MP-DFN model comes at a relatively high computational cost. The diffusion process within the solid electrode particles, typically modeled by Fick’s 2nd law, is the most computationally expensive part of the DFN model. Multiple particle sizes will increase the computational burden, leading to a longer calculation time. Padé approximations have been suggested to replace the commonly used finite difference method (FDM), reducing the computational burden [3]. However, this simplification may not work for larger particle sizes and/or low diffusion coefficients. To address these challenges, a Hybrid-Multiple-Particle DFN (HMP-DFN) model has been developed in this study. The model uses a combination of the FDM and Padé approximation for the various particle sizes in the electrodes. The scaled diffusion length [4] determines whether the FDM or Padé method for a specific particle in the electrode should be applied.A simulative comparison has been performed between MP-DFN, HMP-DFN, and Padé approximation using the fourth order for all particles (PMP-DFN) and the DFN with one particle size using the average value from the particle distribution. In this comparison, the MP-DFN model has been used as a reference as it is considered the most accurate. The main results are shown in Fig.1, which reveal that the DFN model gives the largest simulation errors, especially at the cut-off and relaxation parts. The PMP-DFN shows a relatively good agreement with the MP-DFN and is 60% faster in simulation time. However, relatively large errors can be observed at the beginning of discharge and in the relaxation region, resulting in a root mean square error (RMSE) of 0.4 mV and a maximum error of 8 mV. The HMP-DFN model shows remarkable accuracy in comparison to the MP-DFN model. It neglects the errors from the PMP-DFN model at the initial stage and decreases the error in the relaxation significantly, giving an RMSE below 0.06 mV, a maximum error below 1 mV, and a more than 40% faster simulation time in comparison to the MP-DFN model. Therefore, due to the decreased computation time with respect to the MP-DFN and the low simulation error, the HMP-DFN model is a suitable and practical tool for accurately simulating battery behavior and optimizing battery design.[1] M. Doyle, J. Newman, The use of mathematical modeling in the design of lithium/polymer battery systems, Electrochimica Acta. 40 (1995) 2191–2196, doi:10.1016/0013-4686(95)00162-8.[2] T.L. Kirk, C.P. Please, S.J. Chapman, Physical Modelling of the Slow Voltage Relaxation Phenomenon in Lithium-Ion Batteries, J. Electrochem. Soc. 168 (2021) 060554, doi:10.1149/1945-7111/ac0bf7.[3] G. Fan, K. Pan, M. Marcello, A comparison of model order reduction techniques for electrochemical characterization of Lithium-ion batteries, 54th IEEE Conference on Decision and Control (CDC), Osaka, Japan (2015) 3922-3931, doi:10.1109/CDC.2015.7402829.[4] K. H. Chayambuka, G. Mulder, D. Danilov, P. Notten, A Hybrid Backward Euler Control Volume Method to Solve the Concentration-Dependent Solid-State Diffusion Problem in Battery Modeling, Journal of Applied Mathematics and Physics (2020), doi:10.4236/jamp.2020.86083. Fig. 1. Discharge voltage curves (1.5C) and corresponding relaxation parts simulated with various model types (a); the error curves of the comparison between the HMP-DFN, PMP-DFN, and DFN with the MP-DFN (b). Figure 1

Untangling Complexity in ASEAN Air Traffic Management through Time-Varying Queuing Models

Free route airspace allows airspace users to freely plan a route in en-route airspaces within certain restrictions. It is anticipated to offer the benefit of fuel saving and operational flexibility. Regarding its efficient implementation into the ASEAN airspace, the key challenge would be reducing hotspots with clusters of potential conflicts. This paper designed a time-varying queuing network model, which contributed to untangle trajectory complexity in the most congested hotspot area. A series of fast-time simulation experiments were conducted to identify hotspots in en-route airspace in Singapore FIR. The application of departure time control using time-varying queuing networks successfully reduced up to 45% of potential conflicts. This was achieved within an average delay of 30 min by controlling time of less than 60% of candidate flights. The original contribution of this paper is to develop a novel modeling and simulation framework for composing ideal air traffic patterns. Lastly, we discussed the extension of this study toward a generalized application of the proposed approach in future air traffic management.

Entry Count vs Occupancy Count to assess sector capacity with Fast Time Simulation

An air traffic controller should manage aircraft movements per hour to ensure the safety and efficiency of the traffic system. This study proposed an innovative methodology to estimate the air traffic controllers’ workload through sector occupancy. Starting with a model for the workload of sector controllers designed by Eurocontrol for tower controllers, the authors defined a taskload model from real audio that occurred in an examined Terminal Control Area (TMA). Moreover, trombones have been implemented to sequence arrival flows in the simulated TMA. Therefore, an airside model has been reproduced using Fast Time Simulation software, and its capacity has been calculated in terms of occupancy due to approaching movements. The results permit to evaluate the impact of infrastructural and operative modifications on the sector capacity in terms of hourly entry count and occupancy. Total and maximum occupancy values have been deepened to identify the best predictive factor of controllers’ workload and to manage the airspace configuration.

Vertidrome Airside Level of Service: Performance-Based Evaluation of Vertiport Airside Operations

This paper presents the Vertidrome Airside Level of Service (VALoS) framework, a novel performance metric designed to evaluate airside traffic flow operations at vertidromes in the context of Urban Air Mobility (UAM). As the UAM industry rapidly evolves, the need for a comprehensive evaluation framework becomes increasingly important. The VALoS framework provides a performance-based approach to evaluating vertidrome traffic flow performance, considering metrics like average passenger delay, air taxi in-flight delay, and vertidrome punctuality. Unlike existing Level of Service approaches, the VALoS framework unifies the requirements of various stakeholders, the passenger, the air taxi operator, and the vertidrome operator each with their own performance metric and target. It provides a multi-faceted approach covering airside air and ground traffic flows, arrivals and departures, and performance changes during strategic planning and tactical execution phases. The VALoS is evaluated at 15-min intervals while considering changing stakeholder performance targets and operational uncertainties. For the reference use case, the study demonstrates the significant impact of short-term disruptions, while stochastic deviations can be neglected. Higher traffic volumes due to changing demand/capacity ratios result in higher VALoS variability. The VALoS framework, together with a fast-time simulation, provides a versatile method for exploring future vertidrome traffic flows and supporting strategic vertidrome airside planning and integration. This integrated approach is essential for the evolving UAM vertidrome industry; aligning the interests of different stakeholders and promoting sustainable and efficient vertidrome planning and operation.

Towards a resilience assessment framework for the airport passenger terminal operations

Airport terminals can generate and intensify the undesired phenomena that take place on transportation networks in the wake of disruptive events. The more resilient a terminal is, the more easily it will bounce back to the desired state, causing less nuisance to the users. In this study, we expand the state-of-the-art airport terminal resilience assessment through fast-time simulation and tailored resilience metrics. From the point of view of simulation, we establish delay management strategies based on flight schedule alterations as a response to a disruptive event. In the baseline scenario, no disruption is set, whereas passenger arrival to the airport is blocked for 1 or 2 h in the disrupted scenarios. A previously verified and validated simulation model of a Brazilian domestic airport was used to represent the passenger departing flow, encompassing check-in and security screening. From the resilience perspective, a systematic assessment procedure for new resilience metrics is unveiled, providing means for exploring the metrics’ formulations and their relationship to the operational strategies. Following the simulation runs, resilience metrics for individual passenger and aircraft delays were investigated, something yet scarcely explored in the literature. The results show that the proposed metrics deliver significantly different outcomes from those associated with the current literature. Consequently, we can conclude that a resilience assessment concerning the airport terminal that disregards individual delays may be missing important information. The systematic approach we propose is a contribution in that sense, even though this topic requires more research, considering the vast nature of the possible disruptive events and their subsequent operational impacts.

SINGLE CROSS-BORDER FREE ROUTE AIRSPACE CONCEPT IN THE ASEAN REGION

Single Cross-Border Free Route Airspace (FRA) represents the ultimate long-term idealistic routing rules allowing airspace users freely plan a route in en-route airspace, which is not immediately possible from the administrative and technological point of view. In Single Cross-Border FRA, the whole ASEAN airspace is treated as a single-large FRA where all 12 Flight Information Regions (FIRs) implement FRA, with direct routing between the entry (horizontal entry into FIR or end of Standard Instrument Departure Route, SID) and the exit (horizontal exit from FIR or beginning of Standard Arrival Route, STAR). Single Cross-Border FRA is the most fuel-efficient reference that we could ever reach in the future.All otherhypothetical routing rules fall between the Single Cross-Border FRA and those hypo-thetical routing rules in terms of fuel efficiency. In this paper, we described the potential in fuel efficiency (or benefit achievement scaling) with refer-ence to this ultimate basic fuel requirement and scaled the excess fuel with different routings. We evaluated the fuel burn with a fast-time simulation (FTS) in 12 different scenarios (3 traffic volume levels × 4 routing scenarios). As for the traffic volume scenarios, we simulated 1-day of commercial ASEAN flight schedule data of 3 traffic volume levels (100%, 150%, and 200% of pre-COVID19 traffic volume). As for four routing scenarios with increasing levels of directness: conventional Air Traffic Service route network (ATS) (R1), hypothetical Several-Small FRA route network (R2), Single Cross-Border FRA route network with Danger Areas (R3) and Single Cross-Border FRA network without Danger Areas (R4). Results showed that flight time, fuel burn, and distance travelled were less in R2, R3, and R4 than in R1. There was an average difference of 2.2% in flight time, 1.9% in fuel burn, and 2.2% in distance travelled from R1 to R4.

Impact of Higher Airspace Operations on Air Traffic in Europe

Historically, higher airspace has been used for military exercises and as transit for space vehicles. Riding on commercial space operations’ coattails, more and more vehicles are under development that will make use of higher airspace resources. This will lead to increasing interactions with conventional air traffic since these new vehicles will have to transit through lower airspaces. The management of these operations is necessary to ensure the safe and practicable shared usage of these airspaces. This paper outlines an assessment of the impact of higher airspace operations on conventional air traffic in Europe. Initially, a synthesis of possible use cases was performed, and demand scenarios were developed that served as input to a fast-time simulation. The impact on air traffic was measured by means of flight efficiency parameters. The simulation results showed that the impact is dependent on the type of operation. High-altitude platform system flights and orbital launches cause the largest deviations in flight distance, flight duration and fuel consumption. Higher airspace operation parameters, including location, time, and duration, strongly affect the impact on the conventional air traffic.

Enhancing deep reinforcement learning with integral action to control tokamak safety factor

Recent advances in the use of Artificial Intelligence to control complex systems make it suitable for profile plasma control. In this work, we propose an algorithm based on Deep Reinforcement Learning to control the safety factor profile with a feedback design. For this purpose, we first derive a device-specific control-oriented model with fast simulation time. Then, in order to enhance robustness with respect to external disturbances and model errors, we include an error time integrator into the controller. A cascade of the kinetic and magnetic models with the error time integrator is used in the learning procedure of the feedback controller. Finally, to illustrate the efficiency of the proposed design procedure, the obtained controller is tested in a reference plasma simulator, the Raptor simulator.

RPAS performance model for fast-time simulation research on integration in non-segregated airspace

PurposeThe purpose of this paper was to elaborate the performance model of the remotely piloted aircraft systems (RPAS) which was destined for simulations of the construction characteristics, airspeeds and trajectory of flight in the controlled, non-segregated airspace according to the standard instrument departure and arrival procedures (SIDs and STARs).Design/methodology/approachThis study used systems engineering approach: decomposition of RPAS performance model into components, relations and its connection with components of controlled the airspace system. Fast-time simulations (FTS) method, which included investigation of many scenarios of the system work, minimizing the number of input variables and low computing power demand, is also used.FindingsPerformance envelope of many fixed-wing RPAS was not published. The representative RPAS geometry configuration was feasible to implement. Power unit model and aerodynamic model needed to be accommodated to RPAS category. The range of aircraft minimum drag coefficient differed in the investigated range of take-off mass and wing loading.Research limitations/implicationsFixed-wing RPAS of small and medium categories cover take-off mass (25–450 kg), wing loading (40–900 N/m2) and power loading (8–40 W/N).Practical implicationsThis is a research on integration of the RPAS in the controlled, non-segregated airspace. The results of the work may be used in broadening the knowledge of the RPAS characteristics from the perspective of operators, designers and air traffic services.Originality/valueThe elaborated performance model of the RPAS used the minimum number of three input variables (take-off mass, wing loading and power loading) in identification of the complete RPAS characteristics, i.e. construction features (aerodynamic, propulsion and loads) and flight parameters (airspeeds and flight trajectory).

Switching tube model predictive based controller design for multi‐agent unmanned aerial vehicle system with hybrid topology

AbstractThe decentralized control is provided for the multi‐agent unmanned aerial vehicle system and is designed based on leader‐follower configuration by graph theory. Next, primary model predictive control (MPC) is proposed in order to steer the error signal to the origin and specifically, an optimization algorithm is considered for the primary MPC controller in which the Newton optimization with back‐stepping step size definition is proposed for this manner. A robust modification is investigated for the system based on tube definition and considered to overcome the high‐frequency noise and external disturbances in a bound of 20% of the system output. The high‐frequency noises are designed outside of the primary controller band‐width. The hybrid controller is proposed for the system. The closed‐loop swarm system's control architecture, switches between the primary MPC controller to tube‐MPC controller in the presence of high‐frequency noises and external disturbances. The hybrid controller provides a more energy‐efficient outcome compared to the single mode controller. The final results provides a 7% faster simulation time. Afterward, the flexible formation theory is generalized for system. The possibility of changing leader is investigated in flexible formation and a scenario of switching leader is proposed for the system. The result of tracking for the controller is compared to proportional‐integral‐derivative and the problem of four agents in a flight scenario illustrates the issues and the comparison between the robust controller and the switching Tube‐MPC controller is provided. The results provide that the novel control architecture has a 47.6% lower rise time and 30.8% lower overshoot.

Loop Rerolling for Hardware Decompilation

We introduce the new problem of hardware decompilation . Analogous to software decompilation, hardware decompilation is about analyzing a low-level artifact—in this case a netlist , i.e., a graph of wires and logical gates representing a digital circuit—in order to recover higher-level programming abstractions, and using those abstractions to generate code written in a hardware description language (HDL). The overall problem of hardware decompilation requires a number of pieces. In this paper we focus on one specific piece of the puzzle: a technique we call hardware loop rerolling . Hardware loop rerolling leverages clone detection and program synthesis techniques to identify repeated logic in netlists (such as would be synthesized from loops in the original HDL code) and reroll them into syntactic loops in the recovered HDL code. We evaluate hardware loop rerolling for hardware decompilation over a set of hardware design benchmarks written in the PyRTL HDL and industry standard SystemVerilog. Our implementation identifies and rerolls loops in 52 out of 53 of the netlists in our benchmark suite, and we show three examples of how hardware decompilation can provide concrete benefits: transpilation between HDLs, faster simulation times over netlists (with mean speedup of 6x), and artifact compaction (39% smaller on average).

CO2 and non-CO2 balanced Environmental Scores Module for flight performance evaluation and optimisation

The SESAR2020 exploratory research (ER4) programme CREATE (Grant 890898) developed a climate and weather aware Concept of Operations (ConOps) which encompasses a multi-aircraft 4D trajectory optimisation framework, which utilises a CO2 and non-CO2 balanced Environmental Scores Module (ESM) for the en-route flight phase. The ESM provides a computational method to evaluate the “greenness” of aircraft trajectories. Some components related to the internal ESM scoring are based on expert judgement, which is in line with the technology readiness level (TRL) 1 of the solution. Fast-time simulations were performed to demonstrate the proof-of-concept of the ESM in a multi-aircraft tactical optimisation scenario in the North-Atlantic region. The results show that, because of the simplicity of the metric, the ESM could be well used for trajectory optimisation and tactical replanning, and most likely as well as flight and ATC sector environmental performance evaluations.

SBAS-based navigation system for precision approach in CAT II LVC

Due to lack of advanced navigation systems, secondary airports are normally unable to support instrument approaches in low visibility conditions (LVC) more adverse than Category (CAT) I, making them inaccessible during such conditions. Incoming flights are then delayed, diverted to adjacent airports, or cancelled before take-off. This paper presents an onboard navigation solution, under development in scope of SESAR 2020 Solution PJ02-W2- 17.2, which focuses on technology development to mitigate these traffic disruptions by enabling instrument approaches in CAT II LVC with minimum airport and air navigation service provider investments. The concept builds upon global navigation satellite systems (Global Positioning System L1 frequency – GPS L1, satellite-based augmentation system), which currently enables Localizer Performance with Vertical guidance (LPV) operation down to LPV 200 minima. The designed system includes radio altimeter-based aiding and innovative integrity algorithm to improve navigation performance in support of operations below 200ft down to CAT II / LPV 100 minima. As no such requirements have yet been officially published, authors also defined sets of candidates based on Instrument Landing System Collision Risk Model and other sources. The team developed and verified a Monte-Carlo fast-time simulation toolset to enable representative and statistically significant evaluation of the concept feasibility via its availability throughout Europe, as well as sensitivity to proposed sets of alert limits. Also, flight data collection and a developed software prototype allow for further validations. Simulations demonstrated over 95% average availability with the best results achieving level of 99% depending on used set of requirements. The solution enabling operations in LVC allows unlocking various benefits to business/airline operators (e.g., avoided extra costs, increased predictability), passengers (avoided opportunity loss) and airports (retained profit from flights, less traffic at diversion airports). Technology development also marks an important step towards gate-to-gate autonomous operations.

Scaling multi-agent reinforcement learning to full 11 versus 11 simulated robotic football

Robotic football has long been seen as a grand challenge in artificial intelligence. Despite recent success of learned policies over heuristics and handcrafted rules in general, current teams in the simulated RoboCup football leagues, where autonomous agents compete against each other, still rely on handcrafted strategies with only a few using reinforcement learning directly. This limits a learning agent’s ability to find stronger high-level strategies for the full game. In this paper, we show that it is possible for agents to learn competent football strategies on a full 22 player setting using limited computation resources (one GPU and one CPU), from tabula rasa through self-play. To do this, we build a 2D football simulator with faster simulation times than the RoboCup simulator. We propose various improvements to the standard single-agent PPO training algorithm which help it scale to our multi-agent setting. These improvements include (1) using a policy and critic network with an attention mechanism that scales linearly in the number of agents, (2) sharing networks between agents which allow for faster throughput using batching, and (3) using Polyak averaged opponents, league opponents and freezing the opponent team when necessary. We show through experimental results that stable training in the full 22 player setting is possible. Agents trained in the 22 player setting learn to defeat a variety of handcrafted strategies, and also achieve a higher win rate compared to agents trained in the 4 player setting and evaluated in the full game.