The next major advancement in artificial intelligence may not originate from systems that generate text, code, or images. Instead, it could emerge from AI that comprehends the physical world.
Aparna Pratap, deputy general manager for infotainment at Mercedes-Benz Research & Development India (MBRDI), believes the future of AI in automotive engineering hinges on systems that grasp physics, space, and context far more deeply than current generative AI models.
“In AI research, there are specifically two areas which I follow and which I believe could have immense impact on the automotive side of things, and these relate to AI understanding physics inherently,” she stated during a recent podcast appearance.
She highlighted the work of Yann LeCun on world models and Fei-Fei Li on spatial intelligence. Both research directions aim to provide AI with an intuitive understanding of the three-dimensional world. This is crucial for automobiles, where systems must operate in unpredictable real-world environments.
Today’s neural networks are trained on millions of data points and videos. However, when they encounter edge cases, uncertainty remains. “We are always fingers crossed,” Aparna remarked. If AI progresses toward world modeling and spatial intelligence, “these kinds of predictions become intuitively more intelligent, and also in the real physical space, more natural.”
The implications for engineering could be significant. AI models may require less brute-force data training and more intelligent 3D modeling. This could transform how autonomous and edge-deployed AI systems are built. A model that inherently understands physical space could respond better when deployed in vehicles, robots, or other physical environments.
Mercedes-Benz has already integrated computer vision and AI into vehicle interfaces. Aparna noted that Mercedes worked on gesture control as early as 2018, using computer vision algorithms later deployed in the flagship S-Class in 2020. “You just wave your hand and you’re able to open the sunroof,” she said.
But the industry is moving beyond gesture models. “The world is moving towards voice,” she added, noting that carmakers are exploring new modalities to make user interfaces more intuitive.
Her advice to young engineers is clear: do not chase only the newest tools. Build fundamentals. Core engineering principles, physics, mathematics—especially linear algebra for AI/ML—will remain essential.
She urged students to participate in hackathons, solve real-world problems, publish work on GitHub, and contribute to open-source projects when possible. For agentic AI, she recommended studying open courseware and courses such as Stanford’s CS336: Language modeling from scratch. “The Stanford course is one of my favourites. You learn how generative AI works, what is the infrastructure required, and then how do you start to deploy it for production grade,” she said.
Aparna also recommended a 15-minute daily reading habit. Read anything, and then summarize it without AI assistance.
“In today’s world, the students, the workforce, we all talk to two different species – humans and machines,” she explained. Engineers must learn to prompt machines with precision, but they must not lose the ability to communicate with humans. That exchange of ideas with humans is crucial.
