With the 10th anniversary of Autoware marked earlier this year, attention is now turning to what the next decade should look like for the open-source autonomous driving software pioneered by TIER IV. In this installment of TIER IV People, Microautonomy Division Manager Yutaka Kondo discusses Autoware’s evolution, reflecting on recent improvements, ongoing challenges, and what it will take to establish Autoware as a standard for all forms of autonomous mobility.
— Can you talk about your background and how you got involved with Autoware?
I’ve worked in robotics for more than 20 years. My early research focused on motion generation and human–robot interaction in humanoid robots, and after graduating I worked on software development for various robotic systems. That naturally led to work in mobility and navigation, and gave me exposure to autonomous driving technologies as well.
At my previous company I developed a household robot. As well as being involved in the system engineering, I also worked on business development, PR, marketing and sales. As TIER IV also uses ROS as middleware, I felt my background aligned well with the work being done here. TIER IV’s CTO asked me to take on the challenge of improving the quality of Autoware. A lot of organizations had contributed specific functions, but not many were looking at the architecture as a whole or thinking about how to raise its reliability as a product. That was the position I stepped into.
Before touching the source code, I focused on strengthening the CI/CD pipeline so we had a solid foundation for testing and delivering improvements. Once that was in place, I shifted to improving Autoware itself.
Today, I lead the Microautonomy Division, which oversees research and development not only for autonomous driving software but also for related hardware, system software, operating systems, and even AI accelerators. One of our goals is to make the Autoware release cycle stable enough to integrate directly into products. We’re also working with the Autoware Foundation, the international nonprofit organization that coordinates Autoware’s development, to expand adoption of the platform. AutowareCon, an international developers conference, and the release of Autoware Core through the ROS build farm are part of such efforts.
ℹ️ The two main distributions of Autoware are Autoware Core and Autoware Universe. Core is the maintained, production-oriented subset, while Universe offers a broader set of community-developed features.
— What are the challenges of overseeing a division of about fifty engineers?
One idea I keep in mind comes from High Output Management [a 1983 book by the then-CEO of Intel, Andrew Grove]: A manager’s output is defined by the output of their own team and the teams they work closely with. My role is to help our engineers focus on the right priorities and to create an environment where problems are solved collaboratively rather than in isolation.
Until recently, five departments reported directly to the CTO: Autoware Engineering, AI Accelerator, End-to-End AI Technology, Simulation, and System Software. They worked effectively but often independently. We are now reorganizing them into a unified platform division so that decisions are made with the overall system in mind. To guide that shift, I often refer to the “flywheel” concept: once each part of the system reinforces the next, progress accelerates. Our aim is to build that kind of loop across the division by aligning goals and strengthening the connections between teams.
— What was behind the decision to distribute Autoware through the ROS build farm and what were some of the challenges?
Until recently, Autoware was provided at the source-code level, along with build instructions. In practice, only highly experienced developers could use it. That limited the growth of the community.
By distributing Autoware Core through the ROS build farm, we could package it properly and rely on Ubuntu’s installation system. With setup scripts that download and install everything automatically, the barriers to entry dropped dramatically.
The biggest challenge in making that switch was stabilizing the interfaces. In the past, Autoware’s interfaces changed frequently, which forced users to fix their code constantly. Releasing through the build farm meant Autoware would have many more users, so we needed to ensure nothing would suddenly break.
To address that, we adopted semantic versioning. Users may not always see big changes on the surface, but internally we’re improving the system continuously. Getting that versioning and maintenance process under control took a lot of effort, but it has given us the confidence to release Autoware in a stable form.
One of the biggest challenges before was that building all the source code for Autoware Universe required a high-performance computer, meaning a lot of memory and CPU resources. Now users can instead install prebuilt container images, which significantly lowers the hardware requirements. That’s a huge advantage. The installation and setup process has also become much easier, so I think even junior-high school students could handle it – and I’m pretty sure some already are.
— Now that it has been released, what other challenges remain?
Documentation and tutorials. Users may now know how to install Autoware, but many still don’t know how to set it up or use it in practice. We need to make that easier and more accessible. Another challenge is that Autoware Core and Autoware Universe are still separate. The current release covers only Autoware Core, which has limited functionality for now. We need to gradually bring over more features from Universe and continue developing Core.
Since our mission is to make autonomous driving accessible to all, we need not only more developers but also more users. In that sense, the ROS build farm distribution directly contributes to that goal. Autoware’s GitHub repository has finally exceeded 10,000 stars, so I think it’s gained a great deal of recognition already. But to reach 50,000 or 100,000, we’ll need to look beyond Japan. As we expand globally, it will become harder for us to provide direct support. That’s why things like documentation, tutorials and examples of using Autoware to operate mobile robots or real vehicles will be essential. Sharing those resources openly will help the community grow exponentially.
— As Autoware gains traction globally, what roles do TIER IV and the Autoware Foundation play in expanding the community?
Autonomous driving is needed everywhere, but not every country can afford commercial solutions. Autoware’s open-source model makes it accessible even to regions with limited resources. That’s why we’re seeing activity not only in developed countries but also in parts of Africa and South America.
TIER IV is heavily involved in the Japanese market, particularly in public transportation, logistics, and industrial applications. The Autoware Foundation complements that by building international ecosystems. In Europe especially, AWF works closely with business partners, local governments, and research institutions to expand Autoware’s reach.
At AutowareCon, we saw presentations from all over the world: autonomous vehicles navigating forests, operating on construction sites, and running inside airports. Those real-world examples illustrate how broad the possibilities are. That’s the true power of open source.
— Where do you see Autoware having the greatest real-world impact in the near term?
Most autonomous driving companies focus on privately owned passenger cars. TIER IV’s main target, and where Autoware can have the greatest near-term impact, is public transportation. In many regions, transit demand is rising as populations age and local transportation services are cut back, and this is happening amid a decline in the number of licensed bus drivers. Few companies can address the challenge because it requires close collaboration with local governments and public agencies. TIER IV is in a strong position here because we’ve built those relationships and followed a clear strategy.
In public transport, there is more flexibility: we can use richer sensor configurations, restrict certain lanes to autonomous vehicles, connect traffic signals directly through communication rather than camera detection, and even manage roadside conditions like vegetation. These are advantages unique to this domain. If Autoware becomes the de facto standard for autonomous public transport, it will put TIER IV in a very strong position.
— How do you see Autoware evolving over the next decade?
Autoware has traditionally relied on probabilistic robotics – algorithms that estimate motion while accounting for noise – and rule-based logic. That approach worked well for many years, but as rules accumulated, they began to conflict and created a ceiling for performance.
To move forward, autonomous vehicles need to behave more like human drivers. The current trend involves learning models: large neural networks trained on vast amounts of camera data paired with human steering, acceleration, and braking actions. With tens of thousands or millions of examples, models can replicate human behavior. This paradigm has accelerated rapidly over the last few years, and many major players are already pursuing it. Autoware needs to keep pace. That’s where we’re concentrating our efforts now.
At the same time, learning-based systems have safety challenges: they can’t always explain why they made a decision. That’s where Autoware’s traditional rule-based framework remains essential. We can rely on proven mechanisms to ensure vehicles stop when a person or obstacle appears.
The challenge for the next decade is improving performance in environments that are hard to encode in algorithms – rural roads, unpredictable settings, places without clear structure. Those will require dataset-driven approaches.
— What are the biggest challenges in open-source development and how are you encouraging more collaboration with the global developer community?
Our main bottleneck is the entry point: testing and reviewing pull requests. Right now, the structure is still underdeveloped. We need stronger simulation-based testing and a more active, organized review process. Fixing this is essential for building a healthy cycle: developers contribute, contributions improve Autoware, the product attracts more users, and those users become new contributors. I want to strengthen that entire loop over the coming year.
As for strengthening collaboration, one idea we are considering is a “bounty hunter” system that encourages contributions through monetary rewards. The concept is simple: publish issues on GitHub and offer a bounty to the contributor who solves them.
This approach could give our engineers opportunities to deepen their involvement in Autoware while also expanding their own skills, and at the same time serve as an effective way to attract contributions from developers around the world. It’s an open, dynamic model that could eventually supplement some of the outsourced development we currently rely on.
AutowareCon also has huge potential. If we can expand it into a space where global developers, TIER IV engineers, and the Autoware Foundation can collaborate more closely, it will spark ideas none of us would reach alone.
— What needs to happen for Autoware to approach the ecosystem scale of Linux?
Linux succeeded not only because the kernel was strong but because lots of third-party developers built drivers for every possible device. That diversity made it universal. For Autoware to achieve something similar, we need contributions that support many different sensors, computing platforms, and vehicle types. We’re still far from the Linux level, but I want to move significantly closer within the next year or two. The ROS build farm release was the first major step. If the summit is at 100, we’re at about 10 or 15 today.
The ascent up the mountain will depend on a healthy contributor ecosystem. The incentives for third-party contributors are similar to those of Linux. A contribution can strengthen your career profile. And because autonomous driving is still a young field, some developers contribute simply because they enjoy coding and experimenting with these technologies. That said, our build, execution, and simulation systems are still complex. Making them easier and more flexible is an important part of encouraging more contributions.
— Looking further ahead, what is your long-term vision for Autoware?
People often think of Autoware as software for cars, but autonomous driving can be applied to anything that moves. Once Autoware works reliably for cars, it can extend to small robots, aircraft, planetary rovers, deep-sea vehicles, and more. If Autoware becomes the standard for autonomous mobility, it could support an enormous range of applications.
Because Autoware is open source, anyone can use it, even without major funding. That reflects our vision of making autonomous driving accessible to all. It’s similar to how the Raspberry Pi made computing available to people who couldn’t afford conventional hardware.
In the future, I’d love to see affordable hobby kits, like small autonomous mouse robots powered by Autoware. They could be used in competitions or classrooms and help people learn autonomous driving technologies before becoming professional engineers. That’s one of the greatest strengths of open-source software.
Yutaka Kondo | Microautonomy Division
Yutaka is the manager and principal of the Microautonomy Division, where he steers research and development across Autoware and related technologies. including end-to-end AI, middleware, operating systems, simulation, and AI accelerators. He holds a Ph.D. in engineering. Before joining TIER IV in 2024, he developed autonomous mobile robots at an industrial robot manufacturer, an AI startup, and a consumer robotics company. He is the author of Getting started with ROS 2: Next-generation robot programming and serves on the program committee for ROSCon and the executive committee of ROSCon JP.
TIER IV is always on the lookout for passionate individuals to join our journey. If you share our vision of making autonomous driving accessible to all, get in touch.
We’re currently hiring for the following related positions:
Visit our careers page to view all job openings.
If you’re uncertain about which roles align best with your experience, or if the current job openings don’t quite match your preferences, register your interest here. We’ll get in touch if a role that matches your experience becomes available, and schedule an informal interview.
Inquiries
Social Media
X (Japan/Global) | LinkedIn | Facebook | Instagram | YouTube
More