Decentralized Networks Primed to Advance Autonomous Vehicles

Decentralized intelligence networks could hold the key to making self-driving cars smarter and safer. Many see these vehicles as the future of travel. However, there are still several problems that manufacturers and regulators need to work out.

At their core, self-driving vehicles rely on sensors and other data to stay on course, avoid obstacles, and safely deliver passengers to their destinations. These sensors can include LiDAR, radar, thermal imaging, and other advanced systems. All of these systems create a data footprint that can be used to improve the car’s performance in the future.

Federated Learning

One way that manufacturers have found to improve their vehicles’ self-driving performance is to enable their cars to share this data using a system called federated learning. Federated Learning (FL) is a distributed machine learning approach where individual vehicles train models locally and share only model updates with a central server. This approach preserves data privacy by keeping raw sensor data on the vehicle itself rather than sending it to a central server. The data is used to train a shared model for the ecosystem.

Problems with Federated Learning

The main problem with Federated Learning systems is that you need a direct connection to the server. If you travel a lot in your EV, chances are your internet connection will vary depending on your location. This scenario can lead to your vehicle missing updates.

Decentralized Federated Learning (DFL)

Decentralized Federated Learning (DFL) systems remove the reliance on a central server by enabling vehicles to exchange AI models directly through peer-to-peer communication. These systems can pass their gathered data directly between vehicles. This setup allows for more sharing but still has some problems that could leave a vehicle out of the loop.

For one, your vehicle needs to run into other federated learning network vehicles for the system to function properly. When dealing with large areas or crowded cities, the chances of running into each other become much less, resulting in vehicle data becoming stale and outdated before it’s sent to others. Additionally, these systems only transfer personally gathered data across limited interactions.

Privacy Concerns

Another major issue with decentralized federated learning systems is privacy concerns. When you have a central server, it is easy to determine who is the main person responsible for personal or sensitive data. However, when dealing with a decentralized network, the responsibility falls to the individuals. This structure has led many to worry about privacy breaches or abuses.

Decentralized Intelligence Networks Study

A team of engineers introduces a novel solution to the problems faced by these networks called Cached Decentralized Federated Learning (Cached-DFL). The enhanced vehicle data sharing system was unveiled at this year’s Association for the Advancement of Artificial Intelligence Conference.

The engineers introduced a method of EV data sharing that resembles social media in that every vehicle can pass data between each other freely. The Cached-DFL approach leverages high-speed device-to-device communication, with a range of up to 100 meters under optimal conditions. However, real-world effectiveness depends on factors like vehicle speed, environmental interference, and connectivity stability. Vehicles moving in opposite directions at high speeds may have only a brief window for data exchange. This approach will enhance the vehicle’s ability to prepare for changing road conditions, hazards, and other restrictions.

Cached Decentralized Federated Learning (Cached-DFL)

The Cached Decentralized Federated Learning (Cached-DFL) concept focuses on building a network where intermittent connectivity is expected rather than avoided. As part of this approach, the engineers ensured each vehicle could store and forward data independently when available.

Each vehicle trains its own AI model in this setup. The data from the AI model includes vital details such as road conditions, signals, and obstacles. This data is then automatically passed on to other vehicles when they enter the transmission range.

Multi-hop Transfer Mechanism

Each vehicle acts as a relay in this setup. It stores its data alongside 10 other external models that get passed on between vehicles. It’s vital to mention that the system transfers trained AI models rather than the original data, like its predecessors. This strategy improves performance.

Notably, the vehicles share the most up-to-date AI models when they interact. As part of this approach, all outdated info is eliminated before it can reduce performance. Specifically, the system prioritizes newer AI models over outdated ones, with updates occurring based on vehicle encounters rather than a fixed 20-second interval. Cached-DFL employs a staleness threshold (τmax), typically set at 10 or 20 epochs, to discard obsolete models and ensure relevance in decentralized learning.

Learn from Others

The main advantage of this system is that it allows your vehicle to learn from other vehicles’ encounters. Also, this strategy improves the speed at which relevant data can be shared throughout the network. It takes into account the intermittent state of the network at this time, enabling drivers to access data from beyond their immediate interactions.

Decentralized Intelligence Networks Test

The research team tested their theory utilizing computer simulations. The engineers set up a virtual Manhattan and set their digital EVs to drive its many paths. The vehicles had a speed of 14 meters per second. Interestingly, the simulation was designed so that every vehicle would make a random 50/50 choice at each intersection. This approach allowed each vehicle to create a unique model and share it.

Decentralized Intelligence Networks Test Results

The test results shed some light on how this system can improve EVs in the future. It demonstrated that there are a lot of factors that can play a role in how accurate and timely the data used to create EV models is and how it gets created and shared.

The test showed that the more these vehicles encounter each other, the better the performance. Additionally, it demonstrated that the system was ideal for sharing timely data across large networks of privately controlled EVs. The team noted that key data such as speed, cache size, and model expiration were all found to play a role in learning efficiency.

Decentralized Intelligence Networks Benefits

There are many benefits to the Cached-DFL approach. For one, it’s way more efficient in terms of data propagation. The researchers are able to ensure the majority of the vehicles in their simulations had timely models compared to predecessor systems that could see vehicles running outdated models for weeks.

This system provides a reliable path for self-driving cars to collectively learn and teach each other to be better drivers. Since the method takes into account the fact that your vehicle won’t have 100% uptime, it’s ideally suited for real-world use and could be a valuable tool that manufacturers use to program fleets faster.

Data Sharing

The combination of the benefits of decentralized networks with the ability to share and store up to 10 AI models is a game changer. It allows models to travel indirectly through the network to ensure all vehicles have access to the most vital data. Additionally, the system automatically prioritizes the most relevant info from across varying models derived from different areas, furthering its relevance.

Open Data

Another major benefit of this research is the decision to make the data open for everyone via Github files. This decision will improve innovation and allow other researchers to enhance their findings. Engineers can find examples, tests, cached files, technical reports, and more.

Decentralized Intelligence Networks Applications

There are far-reaching applications for this tech that span past the EV sector. Anywhere you have fleets of autonomous vehicles, this technology is sure to make a difference. Some key areas that the researchers have looked into include drones, robots, and satellites.

Decentralized Intelligence Networks Researchers

The Cached-DFL study was put forth by a team of engineers out of NYU led by Yong Liu,  Xiaoyu Wang. Guojun Xiong, Jian Li and Houwei Cao. Notably, the group received financial support in the form of multiple National Science Foundation grants and the Resilient & Intelligent NextG Systems (RINGS) program.

Companies Leading the Autonomous Vehicle Tech Race

The race to get autonomous vehicles on the roads is in full swing. Manufacturers continue to get closer to fully autonomous vehicles every year. However, this monumental task requires a lot of resources, tech, and a massive network of suppliers. Consequently, there are only a few key players dominating the market currently. Here’s one company that is leading the autonomous vehicle revolution.

While Cached-DFL is still in its early stages, companies experimenting with self-driving technology, such as Uber, may eventually integrate decentralized intelligence networks into their fleets.

Uber (UBER -1.32%) entered the market in 2009 and is based in San Francisco, CA. It was the first decentralized ride-sharing app to gain notoriety. The app’s founders are Oscar Salazar Gaitan, Travis Kalanick, and Garrett Camp. Its vision of a decentralized ride-sharing economy has changed the market forever.

When you think of Uber, your first thought is not a robot taxi. Instead, you probably picture a random person pulling up in their personal vehicle to give you a lift. However, all of that might change in the future because Uber is one of the biggest backers of autonomous vehicle tech, having established and supported partnerships with leading innovators like Waymo.

The company has already put autonomous Ubers on the road for testing in several cities, including their latest venture in Austin, Texas. As part of the approach, the company partnered with Alphabet-owned Waymo, allowing Austin-area Uber clients to upgrade to an autonomous Jaguar I-PACE all-electric SUV for no additional cost.

Today, Uber dominates the ride-sharing market and has branched into other sectors, such as logistics, food delivery, and more. The company currently employs +31,100 people. Many see Uber as a smart addition to any portfolio due to its positioning, history, and innovative spirit.

Latest on Uber Technologies, Inc.

Decentralized Intelligence Networks – Powering Future Travel

Smart cars need smart systems. Consequently, there will be more demand for decentralized intelligence networks in the coming months and years. These systems will allow vehicles to improve their autonomous driving capabilities and enable self-driving cars to enhance performance, venture further, and provide helpful data to others.

Learn about other Machine Learning Projects Now

Study Reference:

^{1. Wang, X., Xiong, G., Cao, H., Li, J., & Liu, Y. (2025). Decentralized federated learning with model caching on mobile agents [Conference paper]. Association for the Advancement of Artificial Intelligence Conference. Retrieved from https://arxiv.org/abs/2408.14001v2}