Research Scientist - Reinforcement Learning

More than 60 percent of new cars sold to feature Level 2 or higher driving automation by 2031. According to a new research report from the analyst firm Berg Insight, 55.6 percent of all sold cars globally fulfilled requirements for SAE level 1 (L1) automated driving and higher levels in 2025. By 2031, the percentage is forecasted to reach 76.9 percent. The percentage of new cars sold fulfilling L2 automated driving capabilities will moreover grow from 35.6 percent in 2025 to 57.3 percent in 2031. As a subset of the L2 category, L2+ provides additional features compared to L2 systems. Berg Insight estimates that 8.0 million new passenger cars sold globally in 2025 were equipped with L2+ ADAS capabilities, corresponding to an attach rate of 9.2 percent. This number is expected to reach 28.4 million units in 2031, corresponding to an attach rate of 31.0 percent. About 4.8 percent of all new cars sold in 2031 are expected to feature L3 capabilities. L4 passenger cars are not expected to scale in meaningful volumes before 2031. Advanced ADAS has become a major differentiator. Today, automakers mainly focus on L1 and L2/L2+ ADAS functions. BMW and Mercedes-Benz have both offered L3 systems, but recently reduced their near-term emphasis on L3 and shifted focus toward more scalable L2+ systems. Chinese OEMs are among the most active in the deployment of sophisticated L2/L2+ ADAS. Leading Chinese OEMs include BYD Auto, Changan, Chery, Geely, GWM, Leapmotor, Li Auto, NIO, SAIC and XPeng. Other OEMs offering sophisticated ADAS include for example Tesla with its Full Self-Driving (Supervised), Ford (BlueCruise), General Motors (Super Cruise), Nissan (ProPILOT 2.0/2.1), Toyota (Teammate Advanced Drive), Hyundai Motor Group (Highway Driving Assist 2), Volkswagen (IQ.DRIVE/Travel Assist) and Audi (Adaptive Driving Assistant Plus). There are a number of categories of suppliers serving the market including Tier 1s, semiconductor solution providers, AD software companies, LiDAR suppliers and map providers. Leading global Tier 1 suppliers include Aptiv, Astemo, Aumovio (formerly Continental), Bosch, Denso, Desay SV, Forvia, HL Klemove, Hyundai Mobis, Jingwei Hirain, Magna International, Valeo and ZF Group. "The automotive semiconductor landscape for ADAS and automated driving includes both vendors of central compute platforms and suppliers focused on enabling semiconductor content", continued Mr. Cederqvist. Leading providers of semiconductor solutions include AMD, Ambarella, Black Sesame Technologies, Horizon Robotics, Infineon, Horizon Robotics, Mobileye, NVIDIA, NXP Semiconductors, Qualcomm, Renesas Electronics, STMicroelectronics and Texas Instruments. ADAS software and integrated driving solution providers such as Deeproute.ai, Huawei, Momenta, QCraft, Wayve, WeRide and Zhuoyu Technology develop automated driving stacks, perception software and turnkey assisted-driving systems for OEMs. There are moreover companies specialising in LiDAR sensors for higher-end ADAS and automated driving applications. Leading LiDAR sensor providers include Hesai Technology, Innoviz, RoboSense and Seyond. Mapping and navigation platform providers such as Amap (AutoNavi), Dynamic Map Platform, HERE Technologies, NavInfo, Mapbox and TomTom provide mapping, navigation and localisation solutions. Latest press releases

Uber's Munich robotaxi and the physical AI turn. Uber is launching a robotaxi service in Munich. Not testing. Launching, in partnership with Israeli AI startup Autobrains Technologies, running on Nvidia's Drive Hyperion platform, with real-time driving decisions handled by multiple AI agents working in parallel. The company chose one of Europe's most congested and regulation-heavy cities for the initiative. The ambition is not subtle. The autonomous vehicle push goes global. Uber is not moving alone. China has set December 2026 as a target for Level 4 autonomous vehicle deployment, meaning full self-driving without human supervision available to the public. Renesas has developed what it describes as the industry's first multi-domain automotive system-on-chip built on a 3-nanometre process, delivering 400 TOPS of AI performance for vehicles that need to make hundreds of decisions per second. SpinQ, the quantum computing company, is working directly with self-driving firm DeepRoute.ai on quantum-accelerated image recognition algorithms designed for autonomous vehicles - one of the first concrete examples of quantum and autonomous vehicle technology converging before either is fully mature. These projects are advancing on separate timelines and in separate geographies, but the underlying logic is consistent. Getting a vehicle to navigate urban traffic reliably at scale requires more computation than current silicon can deliver cheaply. The companies making the most progress are combining purpose-built hardware, AI agents, and computing approaches that were purely experimental two years ago. RECOMMENDED READ The Singularity Is Nearer: When Mylistingo is Website created Merge with AI Ray Kurzweil An updated roadmap for the technologies reshaping human civilisation this decade. Light, quantum, and the chip beneath it all. In early June 2026, researchers reported the creation of a chip that can generate, steer, and read light-based information within a single device, using atomically thin materials and nanoscale structures. The chip exploits what physicists call the "valley" degree of freedom, a quantum property of light that can carry information more efficiently than conventional electrical signals. Practical applications are still years away, but collapsing multiple separate components into one device is typically how chips become cheaper and faster over subsequent generations. Around the same time, QuiX Quantum installed a Feed-Forward Control Unit in its photonic quantum computing stack, enabling real-time adaptive operations within the system. Ultra-fast feedback loops are a core requirement for scaling photonic quantum computers into configurations that can work alongside classical high-performance computing networks. The milestone is unglamorous, but these are the engineering steps that make later breakthroughs possible. Where the capital is going. Gigascale Capital, a new fund founded by former Meta CTO Mike Schroepfer, launched in June with $250 million targeting startups rebuilding physical infrastructure. The fund's thesis is that electrification, AI, reshoring, and climate adaptation are creating investment opportunities in the systems that connect computation to the physical world - power grids, manufacturing processes, logistics networks. Schroepfer spent a decade at Meta overseeing AI research before departing in 2022, and the fund reflects how seriously prominent technologists are now treating physical infrastructure as the next major frontier. The broader capital picture reinforces the trend. Alphabet raised $80 billion in equity. SoftBank committed $52 billion to European data centres. Anthropic filed for an IPO at a near-$1 trillion valuation. Capital is moving toward AI infrastructure at a scale that is beginning to reshape national industrial strategies. The common thread. Robotaxis in Munich, quantum chips built from atomically thin materials, climate capital targeting physical infrastructure, autonomous vehicles aiming for production-level deployment by year end - these are different expressions of the same underlying development: computation becoming capable and affordable enough to leave the data centre and operate in the physical world. The next two years will show which of these systems work at scale and which fall short of their early promise. The experiments running now are the ones that will settle those questions. For more coverage of emerging and future technologies, visit Mylistingo.

DeepRoute.ai showcased its Physical AI advances at the Beijing International Automotive Exhibition, with CEO Maxwell Zhou outlining the company's vision to become "the AI infrastructure of the physical world". Zhou noted that current advanced driving systems, whilst imperfect, already demonstrate safety levels several times higher than human drivers. Chief Scientist Chong Ruan, formerly of DeepSeek, presented the company's Foundation Model, which unifies driving decision-making, scene understanding and behaviour evaluation. The model has reduced the data-driven iteration cycle from five days to 12 hours. DeepRoute.ai reports that over 300,000 mass production vehicles now use its Urban NOA solution, accumulating 1.3 billion kilometres of real-world operation. The company aims to exceed one million units by 2026 whilst improving system safety metrics.