ForzaETH by Autonomous Racing Zürich

Ever wondered how reinforcement learning powers autonomous racing? 🧠🏎️ This talk dives into the techniques that put ForzaETH at the top.

We recently gathered for our general assembly, where we reviewed our achievements, addressed current challenges, and aligned on decisions laying ahead. In the past season we competed in multiple races, including ICRA, IROS and CDC, earning several podium finishes and gaining valuable experience 🏎️. Our research efforts resulted in publications at RA-L and ICRA among others, and we further deepened our involvement in educational activities and outreach. Highlights from the assembly include: – 🗳️ Election of a new board to lead us through the coming season – 🤝 Presentation of our new partners and upcoming team merchandise (stay tuned!) – 🎯 Open discussions on goals, challenges and strategic direction Thanks to everyone who participated and contributed to shaping the future of our team! 🏁

🌟 𝐄𝐓𝐇 𝐇𝐚𝐧𝐠𝐚𝐫 𝐕𝐨𝐢𝐜𝐞 🌟 Autonomous racing is about pushing robotics, AI, and embedded systems to their limits. Enter 𝐅𝟏𝐓𝐄𝐍𝐓𝐇, a global racing league where autonomous 1:10 scale race cars compete head-to-head. ForzaETH by Autonomous Racing Zürich is leading the charge. We spoke with Luca Tognoni (left in the picture, Master’s student in Robotics & President of ForzaETH) and Nicolas Baumann (Right in Picture, PhD researcher & project founder) about the tech behind their success and its impact on self-driving technology. 𝐐: 𝐅𝐨𝐫𝐳𝐚𝐄𝐓𝐇 𝐜𝐨𝐦𝐛𝐢𝐧𝐞𝐬 𝐫𝐨𝐛𝐨𝐭𝐢𝐜𝐬, 𝐀𝐈, 𝐚𝐧𝐝 𝐦𝐨𝐭𝐨𝐫𝐬𝐩𝐨𝐫𝐭𝐬. 𝐖𝐡𝐚𝐭 𝐤𝐧𝐨𝐰𝐥𝐞𝐝𝐠𝐞 𝐟𝐫𝐨𝐦 𝐲𝐨𝐮𝐫 𝐬𝐭𝐮𝐝𝐢𝐞𝐬 𝐡𝐚𝐬 𝐡𝐞𝐥𝐩𝐞𝐝 𝐭𝐡𝐞 𝐭𝐞𝐚𝐦 𝐬𝐮𝐜𝐜𝐞𝐞𝐝? 📘 Luca: "The theoretical background from our studies is crucial. Every race brings new challenges, and understanding control theory, embedded systems, and AI helps us adapt quickly." 𝐐: 𝐇𝐨𝐰 𝐢𝐬 𝐅𝟏𝐓𝐄𝐍𝐓𝐇 𝐝𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐭 𝐟𝐫𝐨𝐦 𝐭𝐫𝐚𝐝𝐢𝐭𝐢𝐨𝐧𝐚𝐥 𝐚𝐮𝐭𝐨𝐧𝐨𝐦𝐨𝐮𝐬 𝐯𝐞𝐡𝐢𝐜𝐥𝐞 𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭? 🔧 Nicolas: "We hit the sweet spot—large enough to develop real autonomous stacks, yet small enough to stay fast, affordable, and safe. Unlike full-scale autonomous vehicles, we can push the limits without worrying about million-dollar budgets, embracing the motto: «Move Fast and Break Things.»" 𝐐: 𝐘𝐨𝐮𝐫 𝐭𝐞𝐚𝐦 𝐰𝐨𝐧 𝐭𝐡𝐞 𝐈𝐂𝐑𝐀 𝐆𝐫𝐚𝐧𝐝 𝐏𝐫𝐢𝐱 𝐢𝐧 𝟐𝟎𝟐𝟑. 𝐖𝐡𝐚𝐭 𝐦𝐚𝐝𝐞 𝐭𝐡𝐞 𝐝𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐜𝐞? 🏆 Nicolas: "We put a lot of effort into refining our control system and had a clear battle plan. At the event, we also held a presentation on our innovation, showcasing the new control system we made. Then, we went on to win the competition—proving that our inventions truly made a difference. After the win, we open-sourced our stack, giving our competitors a little helping hand—just enough to keep things interesting." 𝐐: 𝐖𝐡𝐚𝐭 𝐜𝐚𝐧 𝐢𝐧𝐝𝐮𝐬𝐭𝐫𝐲 𝐥𝐞𝐚𝐫𝐧 𝐟𝐫𝐨𝐦 𝐅𝐨𝐫𝐳𝐚𝐄𝐓𝐇? 🤖 Luca: "The algorithms we develop for F1TENTH closely resemble those used in real self-driving cars. We’ve already tested them in simulations, including the Indy Autonomous Challenge simulation. What sets our work apart is that we can push our systems to their limits, optimizing the robot for high-performance situations—insights that could benefit larger autonomous vehicle projects." 𝐐: 𝐖𝐡𝐚𝐭’𝐬 𝐧𝐞𝐱𝐭 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐭𝐞𝐚𝐦? 🚀 Luca: "We founded the official ForzaETH association just six months ago. Now, we’re building long-term stability and are growing the team!" 𝐇𝐚𝐯𝐞 𝐲𝐨𝐮 𝐞𝐯𝐞𝐫 𝐞𝐦𝐛𝐫𝐚𝐜𝐞𝐝 ‘𝐌𝐨𝐯𝐞 𝐅𝐚𝐬𝐭 𝐚𝐧𝐝 𝐁𝐫𝐞𝐚𝐤 𝐓𝐡𝐢𝐧𝐠𝐬’? 𝐒𝐡𝐚𝐫𝐞 𝐲𝐨𝐮𝐫 𝐬𝐭𝐨𝐫𝐲 𝐛𝐞𝐥𝐨𝐰! 👇🚗 Switzerland Innovation Park Zurich | ETH Zürich | ARIS ForzaETH by Autonomous Racing Zürich | Swissloop Center for Project-Based Learning D-ITET | CELLSIUS aCentauri Solar Racing | ETH Feasibility Lab

Check out our new paper! "Predictive Spliner: Data-Driven Overtaking in Autonomous Racing Using Opponent Trajectory Prediction", published in #IEEE Robotics and Automation Letters (RA-L) and to be presented at #ICRA 2025! Together with Edoardo Ghignone Cheng Hu Benedict Hildisch Tino Hämmerle Alessandro Bettoni Andrea Carron Lei Xie and Michele Magno, we present a high-performance overtaking method broken down into three simple steps: 1) Trajectory Prediction: Utilize Gaussian Process (GP) regression to forecast the opponent’s trajectory, enhancing the accuracy of movement prediction. 2) Collision Prediction: Identify potential collision points (Region of Collision, RoC) by analyzing predicted trajectories, allowing for proactive maneuver planning. 3) Overtaking Trajectory Computation: Calculate an overtaking trajectory that maximizes safety and efficiency, ensuring competitive performance. Explore Our Work: 💻 Paper: https://lnkd.in/d_vU55JT 📽️ Video: https://lnkd.in/dVC-5kY4 🔗 Code: https://lnkd.in/dwE3vy6w Thanks to ForzaETH by Autonomous Racing Zürich, Center for Project-Based Learning D-ITET, The Roboracer Foundation, ETH Hangar, and Switzerland Innovation Park Zurich #AutonomousDriving #robotics #ICRA2025 #RAL

Check out the amazing work done by Onur Dikici, enabling our F1TENTH autonomous racing cars to do on-track system identification.

We wrapped up 2024 with an exciting experience at the The Roboracer Foundation 22nd #F1TENTH Autonomous Grand Prix at #CDC24 in Milano 🏎️🏎️ The competition started strong as we were able to achieve sub-7-second laptimes during the first training session. Our fastest lap clocking in at 6.97 seconds showcases the performance of our Race Stack. A major goal for this event was testing our #ReinforcementLearning controller, which we successfully trained on-track. A key challenge were rapidly changing grip conditions caused by dust on the track, making policy training difficult as it adapts to tire-floor dynamics, thus necessitating frequent tire cleaning. As a result, beating our MAP controller in a racing scenario proved challenging. Despite these hurdles, the event provided valuable insights into the potential of our RL approach in 1:10 scaled autonomous racing. Additionally, a collision with another car led to serious hardware issues breaking the servo and damaging the driveshaft. Our team showed resilience and teamwork, bringing the car back to life in an overnight repair stint. Although we started the following day with a mostly untuned car, we managed to recover and finished in 4th place, missing the podium after a hard-fought small final against UniBo Motorsport - Formula SAE Team University of Bologna. Looking ahead, we are excited to return to the track in May at #ICRA2025, ready to push the limits of autonomous racing once again! A heartfelt thank you to the competition organizers Rahul Mangharam, Ahmad Amine, Federico Gavioli, Francesco Moretti and Enrico Mannocci for hosting such a fantastic event. Explore our Open Source Race Stack: https://lnkd.in/eWjswFXU Our CDC Race Team: Edoardo Ghignone, Cheng Hu, Diego Antolin Garcia Soto, Michael Lötscher

🚗💨 Take a look at our latest video showcasing our advanced overtaking capabilities! Using our Predictive Spliner algorithm, we’re able to plan overtakes across an entire lap by accurately predicting and adapting to the opponent’s racing line. https://lnkd.in/eMtSMjyY

🏎️🏁 Thrilled to have showcased our autonomous racing innovations at the Swiss Robotics Day in Basel last week! Partnering with Center for Project-Based Learning D-ITET, we engaged with a dynamic community of robotics enthusiasts, engineers, and industry leaders, sharing insights into our research and capabilities in autonomous vehicle technology. Events like these are a fantastic reminder of the power of collaboration and curiosity to drive technological advancements. A big thank you to everyone who stopped by to discuss the future of robotics with us!

Last week, the ForzaETH race team participated in the F1Tenth Foundation 21st F1TENTH Autonomous Grand Prix at IROS. 🏎️💨 At the race, we were faced with extremely grippy carpet floor, requiring teams to extract 100% of performance from both their hardware and software. This was in contrast to the robustness tuning the team had been hard at work with, after experiences with slippery racetracks at past races. Faced with this "out of distribution" environment, the team worked hard on-site, leaving no stone unturned as we sought to shave tenths of seconds from our lap times. Our efforts resulted in a 7.14s lap over the roughly 35m track. This is an average speed of ~5m/s (18kph), significant considering the curvature of the track! The car accelerated at 8 m/s² and braked at 20 m/s² – accelerations that mimic full-scale racing. This was sufficient for us to qualify in 2nd place. Unfortunately, in the head-to-head shootout, the team's high-risk, high-reward strategy to minimize laptime bit us as we crashed out of our match which would decide if we would advance to the final. We ended up with a 3rd place finish, behind our competitors Lukasz Sztyber and Scuderia Segfault. Despite this, we still can be proud of our Race Stack, especially in head-to-head mode, where our car demonstrated advanced abilities to estimate, predict, and plan to overtake opponents - made possible by our recent Predictive Spliner publication. We thank the competition organizers Rahul Mangharam, François Pomerleau, Chinmay Samak, Tanmay Samak, and look forward to the next race. Race Blog Post: https://lnkd.in/eAMPKxac Predictive Spliner Paper: https://lnkd.in/eR5CGpEd Open Source Race Stack: https://lnkd.in/eWjswFXU Our IROS Race Team: Neil Reichlin, Tian Yi Lim, Benedict Hildisch, Nadine Imholz, Nicolas Baumann, Edoardo Ghignone, Cheng Hu