New Delhi, June 8 -- The future of autonomous driving will be shaped less by hardware advances and more by the rapid evolution of artificial intelligence, data platforms and organisational capabilities, according to a senior executive at Mercedes-Benz Research and Development India.
As automakers race towards higher levels of vehicle autonomy, the industry is moving away from traditional rule-based engineering and embracing AI-driven development models. This shift is transforming how vehicles are designed, validated, deployed and continuously improved.
"Driving intelligence is undergoing a structural shift, moving from sequential, rule-driven systems to models that learn behaviour more holistically," Parthiv Shah, Senior Vice President, Automated Driving, Base Software, Zone Controller & Wiring Harness at Mercedes-Benz Research and Development India, told TechCircle.
For years, Advanced Driver Assistance Systems (ADAS) depended on carefully engineered algorithms. Radar data was processed into clusters, objects were detected, and functions such as Adaptive Cruise Control were built using predefined software logic. The next phase introduced machine learning-based perception, enabling vehicles to recognise pedestrians, cyclists, lane markings and traffic signs through camera and video inputs.
The industry is now entering a third phase. "The shift is being driven by AI-powered driving models that ingest large-scale, multi-sensor datasets and learn driving behaviour in an integrated manner rather than through separate stages," Shah explained.
The implications extend far beyond the vehicle. Instead of maintaining separate software pipelines for perception, planning and control, automakers are increasingly training large AI models that learn these functions simultaneously from vast datasets. Supporting this approach requires major investments in GPUs, high-performance computing and large-scale data infrastructure.
At the centre of this transformation is data. According to Shah, autonomous driving models today are often 10 to 50 times larger than earlier architectures. However, success depends less on data volume and more on quality, relevance and context."Scale alone is not sufficient, and the data must be curated to ensure it remains context-aware," he said.
As a result, automotive engineering teams are beginning to operate more like AI organisations. Scenario analytics, machine learning operations (MLOps), simulation engineering, logging systems and data traceability are becoming as important as traditional software development. "This shift also moves complexity from algorithm design to data and platform engineering," Shah said. "Teams must now build stronger data curation, scenario analytics, ML Ops, simulation, logging, and traceability capabilities."
The growing reliance on AI is also accelerating investment in virtual validation and simulation. Digital environments allow engineers to generate and test millions of edge cases that may rarely occur on public roads but remain critical for safety.
"Virtual validation and simulation play a key role by enabling teams to deliberately generate and amplify edge cases using synthetic data, something that is difficult, time-consuming, and often unsafe to achieve through real-world testing alone," Shah said.
However, he cautioned against viewing simulation as a replacement for road testing. "Simulation serves as a powerful complement to real-world validation," he said. "At the same time, on-road testing remains indispensable for capturing real-world nuances and validating system behaviour in live environments."
The impact of AI is perhaps most evident in safety engineering. Traditional automotive development relied on deterministic validation and predefined requirements. AI systems introduce a different challenge because their behaviour is influenced by training data, environmental conditions and real-world variability.
"Safety engineering for AI embedded in core driving functions is evolving from a requirements-driven, deterministic mindset toward system-level safety assurance," Shah said.
This is pushing automakers beyond conventional testing towards broader frameworks that incorporate simulation, operational data, runtime monitoring, redundancy mechanisms and controlled fall-back strategies. "Safety shifts from a one-time certification milestone to a continuously managed responsibility aligned with system operation and evolution," he added.
Despite rapid advances in AI and computing, Shah believes the industry's biggest hurdles are no longer purely technological. "As autonomous driving programmes transition to production, the primary obstacles have shifted from pure technology to broader system-level challenges," he said.
The most significant challenge, according to Shah, is organisational readiness. Autonomous driving programmes require companies to move beyond traditional V-cycle engineering models and adopt continuous learning approaches that bring together AI, platform engineering, validation, safety and cybersecurity teams.
"The most critical and underestimated barrier is organisational readiness," he said. "Without evolving into cross-functional, data-centric and security-aware organisations, even strong technology and infrastructure will not translate into safe, scalable and production-ready autonomous driving solutions."
According to Shah, as vehicles become increasingly software-defined and AI-powered, competitive advantage will depend not only on algorithms but also on an organisation's ability to combine data, computing infrastructure, safety governance and operational agility at scale.
Published by HT Digital Content Services with permission from TechCircle.