OPINION Ralf Klädtke he integration of AI into AV simulation has emerged as a game-changer. In the past, AVs were trained over many years. Test drivers logged millions of test kilometers in target environments to create the data required to train an AV’s machine learning (ML) models in both environmental and road conditions as well as pedestrian and other vehicle behavior. It took years of physical testing to validate a system’s capabilities. A real-life setback, such as a collision with a pedestrian, could massively harm a program. With the arrival of generative AI we are now in a new world where virtual environments replicate real-world scenarios with accuracy that was previously not possible, training AVs to navigate highly complex scenarios with ease. One of the key challenges in AV development is creating diverse and unpredictable traffic scenarios for testing. AI steps in by generating realistic traffic models that simulate the chaos of actual roads. Advanced machine learning algorithms analyze vast data sets of actual traffic patterns, enabling the simulation to dynamically adapt and evolve. This ensures that AVs undergo rigorous testing under a spectrum of conditions from routine traffic flow to unexpected events. Experienced human drivers have a ‘sixth sense’. Based on many years of driving, humans can anticipate how other drivers might react in certain situations. Likewise, AI models can predict the behavior of other vehicles, pedestrians and cyclists. AI algorithms can forecast the intentions of other traffic participants and guide the decision making of the AV. Like a human driver, AI-driven simulations continuously learn and train their skills to encounter new scenarios and What do pets and artificial intelligence have in common? They’re both highly trainable – provided the right incentives are offered T “When the model correctly identifies objects and automates the correct response, it gets a reward, reinforcing this behavior” INDEX TO ADVERTISERS Acutronic Switzerland Ltd .................................................... 105 Arctic Falls AB ..........................................................Gatefold, 15 ASC GmbH .................................................................................... 61 ATP – Automotive Testing Papenburg ............................ 99 Automotive Testing Expo North America 2024 ........................................ 17, 19, 20, 21 Automotive Testing Technology Novi Expo 2024 ....................................Inside back cover AVL List GmbH ........................................................................... 55 Bay SensorTec GmbH ............................................................. 46 Chroma ATE Inc. ........................................................................ 36 Colmis AB ..................................................Outside back cover CSM GmbH .................................................................................. 33 Dewesoft d.o.o. ............................................. Inside front cover challenges in traffic scenarios. Compared with a human driver, who requires many years of driving to become exceedingly skilled, AI can be trained in just a fraction of the time by being exposed to thousands of challenges and scenarios in realistic virtual environments without risking any fatalities. Since generative AI scenarios can create extremely accurate representations of the real world, these can be used to train ML algorithms of AVs across millions of situations. Algorithms use physical or simulated cameras, lidar and radar inputs and are trained to recognize the surrounding environment. From there, they create clusters of similar images and automate approved responses to take in certain situations. Depending on a car’s proximity to other infrastructure or people, the model, for example, with the ‘eyes’ of a lidar sensor ‘sees’, depending on the distance, more or fewer dots. It then must conclude what the object is as well as the distance of the object. AI is also used to emulate behavior of sensors such as lidar, camera or radar. Different sensor fusion of these technologies can be simulated and trained. When the model correctly identifies objects and automates the correct response, it gets a reward, reinforcing this behavior. As a result, generative AI can train the algorithms rapidly. Instead of years of testing, AI can achieve incredible progress in training for millions of scenarios within days or weeks. Currently, it is expected that more than 90% of real-world test drives can be generated by generative AI. However, there is still a remaining percentage of situations that cannot be foreseen and cannot be trained in advance. Not only is AI boosting immersive environments and generating complex traffic scenarios to train for adversarial challenges and quick decision making, but it’s also rapidly advancing the capabilities of AVs, paving the way for safer, more efficient and truly autonomous vehicles on our roads. Ralf Klädtke is vice president and CTO of transportation solutions at TE Connectivity. His background includes more than 25 years of technology and leadership experience with global technology companies. As technical leader of the transportation solutions segment, Klädtke is responsible for providing thought leadership and guidance around the future innovation, growth plans and portfolio investments for the organization, as well as in guiding the segment’s technology roadmap Peter Huber Kältemaschinenbau SE ................................ 41 Photron USA ................................................................................ 46 Rototest Europe AB ...................................................................77 S Himmelstein & Co ................................................................. 83 SAB Bröckskes GmbH & Co. KG ..................................... 105 Seoul Industry Engineering Co ........................................... 61 Siemens Industry Software NV .......................................... 52 Spektra Schwingungstechnik und Akustik GmbH .... 41 Stähle GmbH ............................................................................... 96 Stringo AB ..................................................................................... 68 Tectos GmbH ..................................................................................3 Hydrogen Technology Expo Europe ..................... 132, 133 Yokogawa Deutschland GmbH .......................................... 99 ZF Friedrichshafen AG ............................................................ 96 Dewetron ........................................................................................27 DTS ................................................................................................... 80 Durcrete GmbH .........................................................................102 Faist Anlagenbau GmbH ........................................................27 Force Control Industries Inc. ............................................... 80 Hottinger Brüel & Kjær AS ................................................... 30 Humanetics Innovative Solutions Inc. ............................. 45 imc Test & Measurement GmbH ..........................................9 Instron ............................................................................................. 38 Ipetronik GmbH & Co. KG ....................................................102 ITech Electronic Co. Ltd ............................................................ 12 Laba7 ..................................................................................................6 Luxoft GmbH ................................................................................. 71 Messring GmbH ......................................................................... 49 Oswald Elektromotoren GmbH .......................................... 83 136 JUNE 2024 www.AutomotiveTestingTechnologyInternational.com