iTranslated by AI
re:Invent 2025: Leveraging AWS IoT for Catapult Sports' Vector 8 Wearable Devices
Introduction
By transcribing various overseas lectures into Japanese articles, we aim to make hidden, high-quality information more accessible. This is the presentation we are covering in this project, which operates on that very concept!
For transcription articles of re:Invent 2025, please check the information compiled in this Spreadsheet.
📖 re:Invent 2025: AWS re:Invent 2025 - Peak Performance: IoT Innovation in Professional Sports (SPF301)
This video introduces how Catapult Sports leverages AWS IoT services to develop the Vector 8 wearable device, which collects and analyzes athlete performance data. It is deployed across 128 countries, in over 40 sports, and by more than 5,000 elite teams, generating 600 types of metrics from 10Hz live data and 100Hz raw data. By utilizing AWS IoT Core, Greengrass, and Shadows, they significantly reduced device onboarding time and improved firmware update speeds by 32 times. Unique metrics like 'Player Load' enable injury prevention and performance optimization, while integration with video allows for tactical analysis. The system has the processing power to collect 36 million data points during a 2.5-hour NHL game and upload them to S3 within 30 seconds.
Please note that this article was automatically generated while maintaining the content of the existing lecture as much as possible. There is a possibility of typos or incorrect content.
Main Content
A New Era of Sports Technology: Introducing Catapult Sports
I usually like to start sessions like this a bit interactively, so could you bear with me? Please raise your hand if you knew that almost every professional or semi-professional football team, pro team, or individual athlete uses some kind of wearable device to track data during games or practices. There are quite a few more people raising their hands than I expected. You might have thought about what type of data they are capturing or what analytics they can generate from that data, but you might not have thought about the companies actually manufacturing these devices or the market they serve. That is what we are going to talk about today. For those who didn't raise your hands, I hope that today you can see another side of sports—sports supported by data.
My name is Mike. I am a Sports Solutions Architect based in Australia, and joining me is Greg, our Senior IoT Specialist Solutions Architect. Today, I'm going to talk about one of my customers, Catapult Sports. I will refer to them as Catapult for the rest of the presentation. I will talk about who they are, what they do, and how they are transforming sports.
We will talk about athlete data ingestion—what they capture and how they do it. Then, after that data is captured, we will cover what you can do with it: what types of analytics, metrics, and athlete performance improvements are possible. We will dive a bit deeper into the device software architecture and how they used AWS to build these wearable devices. We will discuss device onboarding, how they manage updates, and how they roll out different types of settings. Finally, we will look toward the future and what Catapult can do next to continue evolving data in the world of sports.
Now that I've mentioned Catapult, let me introduce who they are. Catapult exists to unlock the potential of every athlete and team around the world. Operating at the intersection of sports science and analytics, Catapult's products are designed to optimize performance, improve resilience, and quantify return-to-play. The company is listed on the Australian Securities Exchange and has over 400 staff across 24 different locations worldwide. Instead of just explaining who they are and what they do throughout the presentation, why don't we watch a short video to set the scene and provide some context?
Athlete Monitoring Solutions and the Full Picture of Vector 8
I intend to mention a few things you might have seen in that video. Catapult has several different solutions, and while we don't have time to cover everything, I'll touch on three and dive deep into one of them today. First, there is athlete monitoring. You might have seen the small device inserted into the vest worn by the player in the video. This is designed to optimize performance, reduce injury risk, and improve recovery. This is the content we will be diving deep into today.
The second solution is scouting and recruiting. This is designed to help teams find, evaluate, and acquire the right players. And finally, to bring context to data, Catapult provides a video suite offering, which connects data to video to transform strategy, performance, and training. Before diving deep into the athlete monitoring solution, let's take a look at Catapult's market. Catapult operates in over 128 different countries. They have customers in over 40 different sports, and 5,000+ elite teams are customers. Catapult's customers include some of the world's largest clubs and franchises, from the NFL to the NHL and the English Premier League. This is an example of the top sports where Catapult has customers.
Now, let's dive deeper into the athlete monitoring capabilities provided by Catapult. Catapult tracks comprehensive performance metrics, providing coaches with actionable insights for training and competition. I would like to specifically mention a few of these metrics. "Player Load" is one we will dive into more later, but it quantifies overall physical stress. This is a key tool for workload management and injury prevention throughout the season. Next, there are "Speed Zones." Speed Zones categorize an athlete's output across speed thresholds, helping to understand the demands of the game and prescribe appropriate training intensity. For those involved in contact sports, "Contact Intensity" measures collision forces in these sports. This is crucial for managing player safety and recovery needs. Finally, and very importantly, Catapult can identify sport-specific movements like jumps, tackles, and throws. These are automatically detected and quantified, providing tailor-made insights for the 40+ sports mentioned earlier. By combining these metrics, data-driven decision-making becomes possible, keeping athletes healthy and ready for competition while optimizing performance.
I would like to introduce the Vector 8. The Vector 8 is Catapult's next-generation athlete monitoring solution designed for both elite and competitive sports environments. The Vector 8 consists of four key components. First, there is the Vector 8 Tag, which was shown in the video. It is compact, durable, and packed with advanced features. It uses GNSS and local positioning for accurate location tracking, working seamlessly both indoors and outdoors. It captures sport-specific events, monitors heart rate, processes data on-board, and supports integration with third-party Bluetooth and Ultra-Wideband (UWB) sensors. It provides up to 6 hours of battery life. For those unfamiliar, Ultra-Wideband, or UWB, is a wireless communication technology that uses short-range, low-power radio signals to enable precise location tracking and secure data transfer.
It is important that each athlete has their own Vector 8 Tag. The Vector 8 Dock is used after a session or match, where the tags are returned to the dock. The Vector 8 Dock enables high-speed WiFi downloads and effortless data synchronization directly to the cloud. It supports simple workflows, local or remote device management, and with on-board computing, it can be used anywhere without the need for additional equipment. Next, there is the Vector 8 Receiver. This connects via both WiFi and Ethernet, streaming live data to the cloud for real-time analysis. It supports both online and offline workflows, providing real-time diagnostics for performance and troubleshooting. Finally, there is the Vector 8 Relay. This extends coverage across larger venues, eliminating the need for multiple receivers. With a compact form factor, it boasts a 400-meter range, making scalable deployments efficient and cost-effective. By combining these components, the Vector 8 becomes a powerful suite for collecting, streaming, and analyzing athlete data, driving decision-making for both training and games.
Data Ingestion Architecture Powered by AWS IoT Core
Now, I would like to talk about how they achieve this. Here, I introduce the first AWS service of the day: AWS IoT Core. AWS IoT Core is a managed service that provides scalable, high-availability endpoints in the cloud. This means you do not need to create endpoints for each individual device. It provides device authentication and authorization, encryption in transit, and large-scale device management, which is critical for Catapult's use case. Most importantly, there is the Rules Engine. The Rules Engine is how you connect the data coming through IoT Core to applications. The Rules Engine serves as a bridge between IoT Core and applications, managing large device fleets ranging from thousands to tens of thousands of devices. This is why it is an excellent service supporting Catapult's Vector 8 technology, and Greg will explain in more detail later how much they use it.
Before diving into detail on how Catapult achieves this, I want to talk about the different ingestion paths and the various types of data Catapult can collect. First, there is device health. This is the battery status and firmware version of the Vector 8 Tag itself. Next, there is hot data. This is live data sampled at 10 Hz during a game or practice. This includes acceleration, speed, player positioning, heart rate monitoring, and so on.
Finally, there is cold data. This is raw inertial data sampled at 100 Hz after a game or practice. This forms the foundation for the machine learning inference that Catapult performs. They do not stream this through hot data; instead, they run a batch approach against the cold data. But how is this realized?
Let's start with device health during a game or practice. This slide illustrates the real-time device health monitoring architecture, a critical operational feature for maintaining fleet reliability during live games and practices. It begins with the player wearing the Vector 8 Tag, and the captured data is transmitted to the Catapult receiver via UWB. Since the receiver is on the sideline, low latency and high-precision tracking are achieved.
From the receiver, device telemetry is published to AWS IoT Core via MQTT. This creates a secure, scalable ingestion pipeline capable of processing data from hundreds of devices simultaneously. The AWS IoT Rules Engine processes the incoming messages and routes them to Amazon MSK (Managed Streaming for Apache Kafka) for real-time streaming and buffering, ensuring that important device health data is never lost. From MSK, the data flows into Amazon Elastic Container Service (ECS), where Catapult's containerized analytics services run.
This processes battery levels, firmware versions, and other diagnostic information in real-time. Results are published via Amazon API Gateway and delivered to the Catapult web interface. Here, coaches and equipment managers can monitor device status at a glance. An example payload is shown on the screen, where you can see battery and firmware versions. This enables proactive responses before a device fails at a critical moment.
Before implementing AWS IoT, Catapult lacked visibility into device status, health checks, and usage. Now, they have real-time insight into every single device. This AWS-powered architecture realizes high availability, scalability, and real-time visibility across the entire device fleet, minimizing downtime and maximizing performance reliability. Now, let's look at hot data. As mentioned earlier, this architecture delivers real-time athlete performance data at 10 Hz—10 updates per second. This allows coaches to make instant decisions during games and practices.
The tags are equipped with inertial sensors for precise positioning and tracking, as mentioned. Tags transmit to receivers via UWB, after which data is streamed to mobile devices and web applications. The payload is updated in real-time, and here is a small example. As you can see, we have heart rate, position, speed, and acceleration. Coaches can see instant visualizations on their iOS apps, allowing them to compare player workloads, check positioning heatmaps, and see individual metrics on these devices.
This enables on-the-spot interventions: substituting fatigued players, adjusting tactics based on movement patterns, and preventing injuries. Finally, there is cold data. This is sampled at 100 Hz after a game or practice, and it is what Catapult uses to perform machine learning inference. To put numbers to it, a two-hour session with 30 athletes can be uploaded to S3 in under 30 seconds.
Imagine an NHL game. With 20 athletes on each side, that is 40 athletes in total, all wearing Vector 8 tags, with inertial data sampled at 100 Hz. In that 2.5-hour game, 900,000 data points are captured per athlete—a total of 36 million data points for the entire game. We have discussed data ingestion, what it looks like, and how it is performed. Now, I want to talk about what happens after that data is ingested.
Insights from Data: Player Load and Effort Breakdown
I will mention a few metrics, but because Catapult can create 600 unique metrics per player per game, we do not have time to cover them all. On the screen, I am showing different sports and some of the sport-specific metrics that can be captured for each. Vector 8 provides AI-driven, sport-specific live data so that coaches can monitor athlete output in any environment.
This is essentially automated performance insight. Catapult runs machine learning algorithms over raw data, processing large datasets quickly and accurately to identify trends or patterns that might be missed by the human eye. These sport-specific movements are extremely important, but I want to think from the perspective of a coach. I have several clients in Australia, and before they used Catapult, they were spending over eight hours on weekends watching video and coding events. After adopting Catapult, they were able to almost halve that time because they could automatically identify these events through the tags rather than the video. I will talk more later about what happens when you cross-reference the data collected from these devices with video.
Now, one of the metrics I mentioned earlier is player load, and you might be wondering why I am in a math class. If you have done math before, the minus sign should mean subtraction, not a subscript. So, let’s define player load and put it into context. Player load is the sum of accelerations across all axes of the internal tri-axial accelerometer during movement. It was originally developed by the Australian Institute of Sport, with rugby union as the intended customer. It is important to know that this output is just an arbitrary unit. It is a volume measurement. In other words, it is how much work the player did.
We can use this to compare players against each other, or to compare a player's workload over time to see how they are recovering from an injury or if they are tracking toward an injury. It has advantages compared to using effort metrics like distance, because it can be accumulated during tackles, ruck work, non-running activities, or activities away from the ball. To put it into context, let's use an example. I love tennis and play it often. Suppose I am recovering from an injury and the player load for a session is 200. This means the effort for that session was 200. The coach is satisfied, I am satisfied, and the performance looked quite good.
In the next session, the player load is 300. This means I worked 50 percent harder than the previous session. However, the output dropped slightly. This identifies two things: I worked harder, but the performance was worse. Perhaps I was not ready to return from the injury, or I might have aggravated something. I want to be very clear here: this is not a definitive metric. It is something that can be used as an input to other statistics. It can be used as an input as a rhetorical metric. How did you feel during the game? Or it can be used in other statistics.
One of my clients in Australia started implementing this, and they noticed that the player load was consistently high, so they decided to delay the player's return by a few weeks. They had been returning from injury early, but their output was significantly less than usual. This is a volume measurement, an example of how much effort a player spent in a session or game. Speaking of effort, Catapult provides effort breakdown reports, which consist of three main sections. There are effort profiles, which track each acceleration or velocity event for each player—we will see what this looks like in a moment. Next, there is effort location mapping, which is a visual representation showing exactly where on the field or pitch the effort occurred, using soccer in the next example. And finally, there are velocity traces.
Coaches can visualize how speed increases and decreases in each individual athlete's effort. Let’s take a look at what this looks like. There is a lot of information on the screen, but first, I want to draw your attention to the area in the bottom left. This is the detailed profile of all efforts. Each acceleration or velocity event is analyzed by duration, peak intensity, and recovery time, helping to understand the physiological load and tactical purpose of each effort. On the right, a soccer pitch is displayed. This is the effort location mapping. Coaches can see exactly where on the field each effort occurred.
This gives context to high-speed movements and also to the effort profiles on the left. Finally, there are high-resolution velocity traces. By accessing 10Hz raw velocity traces, practitioners can clearly see how speed increases and decreases across each effort. This supports deeper quality control and a better understanding of player output.
Effort breakdown provides coaches with tools to move from assumptions to precision. By turning raw movement into actionable insights, teams can design smarter drills, manage workloads with greater control, and ensure that players are executing exactly what the game demands. You may have seen a screen like this playing in the video earlier. The important thing is that wearable data without any context or video is essentially just noise. You can use it for analytics, but it cannot truly explain what actually happened on the pitch or field.
Vector 8 automatically pushes athlete performance data to Catapult’s Pro Video suite. This allows users to match physical output with what actually happened on the pitch or field. Coaches and analysts can immediately review, tag, and describe moments of interest, improving athlete understanding and engagement. You can capture, synchronize, and live-stream multi-angle video from any device, anywhere. This is how they bring new context to team and player performance by integrating player performance data with video.
Integrating video and wearable athlete data is changing how people think about sports. It supports decision-making, quantifies recovery, and hopefully, in the future, will enable injury prevention. We have talked about Catapult, the data they can capture, and the types of metrics that can be created on top of that. Now, let’s take a deeper look at how they are utilizing AWS.
Device Software Architecture with AWS IoT Greengrass
Let’s dive a bit deeper into how Catapult uses AWS IoT services to build the Vector 8 product, the benefits it brings, and the value it is creating for customers. First, let’s look at the device software architecture. AWS IoT Greengrass is the service that Catapult uses as the foundation for the Vector 8 dock and receiver. Greengrass consists of two things: an edge runtime and a cloud service, which help build, deploy, and manage multi-process IoT applications at scale. The edge runtime is open source and available on GitHub.
The processes or services managed using Greengrass are modeled as components. These components are executables, binaries, scripts, container images, or almost any code that can run on a device. They can be executed under the control of Greengrass. Once you have a multi-process application, dependencies arise. Greengrass supports Catapult by supporting component versioning and component dependency resolution. It also supports logs for diagnostics. There is a large collection of pre-built AWS components for solving common challenges. And there is inter-process communication, IPC for sending messages between components and processes, as well as access to some advanced Greengrass features.
Looking at the software architecture at a very high level, the Vector 8 dock and receiver architecture has an embedded Linux operating system at the bottom. The Greengrass edge runtime sits right on top of it, managing multi-process applications. This application consists of custom Catapult components supplemented by components provided by AWS. Moving from the general to the specific, let's look inside the receiver software architecture. Regarding the data ingestion we saw earlier, the tags communicate wirelessly with the receiver using UWB. Inside the receiver software, there is a data processor custom component developed by Catapult, which receives device health data and live data, including 10Hz in-game data.
The data processor component uses Greengrass's inter-process communication feature to publish device health messages to AWS IoT Core via the Greengrass nucleus. The Greengrass nucleus is the heart of the Greengrass edge runtime. Greengrass includes a message spooler, which is convenient when connectivity is lost. If a dock loses its internet connection, Greengrass will buffer those device health messages and send them later when connectivity is restored. This eliminates the need for Catapult to build undifferentiated heavy lifting just for buffering and retrying on their own.
In addition, Catapult has custom components that deploy a database, a RESTful API, and a web server, which are used to serve 10Hz live in-game or practice session data to the iOS app on the sideline—the app used by coaches and sports scientists. Moving to the dock, as you heard earlier, when tags are used in the dock, they are returned to the dock not just for charging, but to retrieve data from the tags. On the dock, there is a custom component called a retriever component that can retrieve cold data, i.e., raw 100Hz post-game data, as data files, and write them to the disk on the dock.
Then, the raw file upload custom component takes those data files and publishes them to an Amazon S3 bucket. As you heard from Mike, this entire process of getting all the data for all athletes from a single game or training session is usually completed in just tens of seconds. Device onboarding, or the initial setup of a device, is a really big topic for Catapult. With Vector 8, the goal was for onboarding to be a self-service experience, where customers could go from unboxing to athlete tracking within 10 minutes.
To understand why we set such a goal, let's go back a little and explain how the initial setup worked for the previous generation product, the Vector 7. The Vector 7 was not a connected product; it was not an Internet of Things device. When Catapult customers received a Vector 7, they had to unbox the device and install a desktop application on their laptop. After that, they had to schedule an appointment with Catapult customer support, wait for the time of the appointment, and then work with customer support to configure the device, apply firmware updates that had been released since the device was shipped, and then test to ensure the device was working properly.
That was a long process, usually taking about two days, at least a lot of time, and was very manual for both Catapult customer support and Catapult customers. Now, with Vector 8, the initial setup process is similar to what you are accustomed to when setting up consumer electronic devices. Customers unbox the device, install the Catapult app on their smartphone, and register the device within that app. This also provisions Wi-Fi credentials to the device. After that, the device can connect to the cloud, automatically download settings and firmware updates, and finally publish its status. From start to finish, it is a 10-minute process without the involvement of Catapult customer support, creating a much more seamless experience for customers. In other words, it is good for both Catapult and their customers.
In terms of the value AWS IoT provides, it helps with scaling and automating device provisioning, configuration, updates, and status reporting. I mentioned device updates, and there are many details there, so let's elaborate. Security is paramount at AWS, but in the IoT domain, I would say over-the-air (OTA) updates are "feature zero." This is the foundation upon which everything else can be built. With OTA updates, you can easily add new features and enhancements to a device even after it has been shipped to the field. You can easily patch devices at scale, resolve bugs and customer support issues, and help prevent other customers from reporting the same problems. OTA updates also help realize innovation by allowing you to iterate on device software more quickly and build the features customers desire based on their feedback.
As you saw earlier, the docks and receivers use AWS IoT Greengrass. Greengrass supports OTA updates using a feature called Greengrass deployments, which is available out of the box. Greengrass deployments allow Catapult to define a set of components to be deployed to a dock or receiver. Some components are shared between the two, but they each also have their own components because they do different things. Using these deployments, you can target a single device, but you can also target a group of devices, so deployments work at scale by deploying to a collection of a large number of devices.
This is resilient to intermittent connectivity. When Greengrass receives a deployment, it needs to pull down those component artifacts—program binaries, executables, and container images. If there is any interruption in the connection, Greengrass handles all the retries, pauses, and resumes to get all the artifacts to the device. This eliminates the need for Catapult to build that boilerplate code themselves. The updates are fine-grained; Greengrass calculates the difference between the current state and the state required by the deployment, and only downloads what is necessary to match the state specified in the deployment.
Greengrass has rollout and rollback settings. If Catapult performs a deployment containing a faulty component that fails to start and enters a corrupted state, Greengrass automatically reverts to the previous deployment state, leaving the dock or receiver in a good working state. Again, Catapult does not need to build this functionality themselves. Finally, it is integrated with operating system updates. Catapult can use Greengrass deployments to patch, update, and manage the embedded Linux operating system.
Optimizing Operations with Over-the-Air Updates and Configuration Management
For docks and receivers, it is easy because they use what Greengrass provides out of the box. For tags, it is a bit more complex. The tags run FreeRTOS and are microcontroller-based devices, but their UWB interface is a custom protocol, and so is the Wi-Fi protocol used when the devices are in the dock. As a result, we needed something a bit more customized for tag updates.
Catapult starts by creating a Greengrass deployment that transfers the tag firmware binary from the cloud to the dock. Then, they use the AWS IoT Jobs feature to trigger the update process. The device manager custom component subscribes to the necessary MQTT topics to be notified when a job arrives, and performs that subscription via the IPC feature provided by Greengrass. Then, when the tags are returned to the dock, the device manager custom component distributes that tag firmware binary to each tag.
This is extremely powerful for Catapult. Because after they create the deployment in the cloud, the next time the tags are returned to the dock, they are updated. The really significant benefit Catapult is getting from AWS IoT and over-the-air updates is that, based on their own data, they are 32 times faster than their previous Vector 7 product generation. A big reason for the acceleration is that previously, with Vector 7, the customer was in the loop. They had to use a laptop and a desktop application to update the tags. Now it is an automated process. As soon as the tags are returned to the dock, they are updated.
Another benefit arising for the same reason is that essentially almost all tags in the fleet are always running the latest firmware. When a tag is returned to the dock after an update is published, it gets the latest firmware.
Tags not running the latest firmware are only a small percentage that for some reason have not been returned to the dock—e.g., they are lost or misplaced. This aspect is really powerful because with their previous product generation, according to Catapult's own historical data, more than 50% of customer support issues were simply resolved by updating the device to the latest firmware. Now, with Vector 8, this happens automatically, and the number of customer support cases has decreased significantly. This is clearly good for both Catapult and their customers.
Over-the-air updates also support innovation. If you want to innovate, it helps to be able to iterate quickly on ideation, execution, and delivery, and to have an efficient development process. Usually, you ideate and plan, build new value, and deploy new increments. From an IoT perspective, that means putting it on physical devices. You test, measure, learn, gather customer feedback, and then start the cycle anew. The faster you spin this flywheel, the more innovation is possible.
As you can see, Catapult is now much faster in the deployment phase with Vector 8. With a 97% penetration rate, when they perform a deployment, it reaches a much larger number of customers much more quickly. More feedback, more learning, and more testing are happening. One half of this flywheel is performing much better thanks to Vector 8's over-the-air updates, which puts pressure on the other two parts to speed up as well. In particular, it is putting pressure on Catapult's firmware team, changing the way they work. They switched from a kanban process to sprints, and the release cadence has shifted from monthly to weekly, and in some cases even daily, especially when developing and beta-testing new features.
When working with beta customers, they can iterate very quickly and help develop the features customers actually want. This is good for both Catapult and their customers. All wearable devices have quite a large number of configuration parameters, and now, with the cloud connectivity available to Catapult, they want to be able to configure the tags from the cloud as well, without needing an application at the edge. Settings include individual athlete settings, metric settings related to sport-specific movements, and specific requirements for different teams, different coaches, and different sports scientists. It also includes radio frequency settings that help tune the wireless communication between tags and receivers in different environments.
Catapult uses the AWS IoT Shadows feature to assist with configuration. Shadows are JSON documents that help AWS IoT synchronize between the cloud and devices. In Greengrass, there is an AWS-provided component called Shadow Manager, which implements the device-side portion of this feature. You can get shadow functionality on your device right out of the box, which is what Catapult did in the dock.
When a Vector Web user changes a setting, Shadow Manager recognizes that the shadow document in the cloud has changed. It is automatically synchronized to the local shadow database. The device manager custom component can use the inter-process communication feature to retrieve the shadow document, and the next time the tag is returned to the dock, it automatically receives the latest settings. Like wireless updates, when you make a change in the cloud, the tag receives that change the next time it is returned to the dock.
The value for Catapult is that this has become very loosely coupled. Even if the dock is offline or the tag is in use, they can safely make configuration changes in the cloud, and the tag will receive the latest settings the next time it is returned to the dock. Customers are no longer in the loop. Previously, for configuration updates, customers had to be involved using a laptop and a desktop application. Now they are out of the loop, reducing their burden. You can also make configuration changes to some of the configuration parameters in the cloud. This is good for both Catapult and their customers.
Remote Diagnostics and Future Outlook: Edge ML and AI Analytics
Device diagnostics. Going back to the Vector 7 product generation, Catapult relied heavily on TeamViewer to reach devices for diagnostic purposes. They had to ask customers to run TeamViewer on their laptops so that customer support representatives could have remote sessions. Now, with AWS IoT, they have the ability to access devices more directly.
Secure tunneling is an AWS IoT feature that Catapult uses for remote SSH sessions to docks and receivers. There is an AWS-provided component, the secure tunneling component, which makes implementation on the device simple. Catapult uses this on docks and receivers, and it is included in Greengrass deployments. When a customer support engineer needs a remote session with a dock or receiver, they go to the AWS console to create a tunnel.
A notification containing a secure access token is sent, and the secure tunneling component creates a tunnel between the cloud and the SSH daemon running on the dock or receiver. The customer support engineer can then establish an SSH session. Previously, customer involvement and TeamViewer were required. We no longer need appointments with customers, and we no longer burden them. They only need to agree to Catapult accessing the device.
Catapult can easily access the devices, but secure tunneling does not require inbound open ports. It creates connections using outbound port 443 and temporary inbound ports, which works even behind firewalls. All of this combined is good for both Catapult and their customers. Finally, regarding Log Manager. Log Manager is also an AWS-provided component. Previously, when Catapult needed to access device logs with Vector 7, they again relied on TeamViewer and customer involvement.
Now, by using Log Manager, they can retrieve log files from custom components and Greengrass log files and publish them to Amazon CloudWatch. Customer support can access these device logs in CloudWatch at any time. That is, near real-time logs are always available in the cloud, which Catapult can access at any time. Again, there is no need for appointments with customers, and no burden is placed on the customer when Catapult needs device logs. This is good for both Catapult and their customers.
The final slide is about the future outlook. We have covered many of the benefits and features Catapult gets from having devices connected to the cloud and becoming IoT devices, but there might be other things they can do. As you saw earlier, the 10Hz live data stream is only sent from the device to the iOS app on the sideline, but Catapult is interested in streaming that data live to the cloud. This could be used for in-game analytics in the cloud during matches or practice sessions, but also enables remote access for coaches and sports scientists who are not on the sideline.
As Mike showed earlier, ML inference is performed post-processing after games or practice sessions, and was executed in the cloud. Catapult does some minor ML inference on the devices, but not much; they are interested in shifting more machine learning inference to the edge, either to the receivers or the tags. By doing so, they will be able to perform these insights during games or practice sessions. Part of the reason they are interested in this direction now is that the OTA update mechanism makes this much more achievable. With the previous Vector 7, if inference models were at the edge on the device, Catapult didn't really have a good way to update them, iterate, enhance, or innovate over time.
Now, because they have OTA updates, they can consider deploying more models to the edge because they will be able to manage, iterate, and enhance them. In other words, they can have a true ML Ops pipeline that deploys all the way to the device. They are also looking into AI analysis of tag data across the team—about team movements, how players move as a group, and team tactics. They are also looking at AI for natural language search of video, such as multimodal embedding models, which are very powerful for that use case.
That's it. We have reached the end. Mike and I would like to thank you very much for your attendance and attention. We would appreciate it if you could provide feedback in the app, and we would be really grateful if you could take the time to do so. Also, if you would like to connect with us on LinkedIn, we would be happy to do so.
Discussion