Advanced driver assistance systems (ADAS) have already reset baseline expectations for safety on the road. With global road traffic deaths still reaching approximately 1.19 million per year, the push to automate safety is urgent. ADAS features such as lane keeping and automatic emergency braking are increasingly treated as basic protections rather than luxury add-ons. More than 58% of new passenger cars produced in 2024 globally now include at least three ADAS functions as standard equipment, signaling that electronic perception has become part of the safety baseline.
However, Level 4 automation sits on a different tier of responsibility. While ADAS (Level 2 or lower) reduces the probability of human error, L4 removes the human fallback entirely, and the engineers building it are quietly rewriting what safety, dependability, and fleet operations look like. Nishant Bhanot, Senior Sensing Systems Engineer at Waymo, has spent the past several years architecting perception and sensing stacks for self-driving programs, including serving as an Associate Editor on the SARC editorial board for the Journal of Innovative Science. His core approach is straightforward – treat redundancy, fault coverage, and downtime as first-class design constraints so every architectural decision shows up in fleet availability, safety margins, and program economics, not just in lab benchmarks.
From Driver Assistance To System Responsibility
As ADAS becomes the norm, the line between “assistance” and “automation” is shifting from comfort features to system-level responsibility. Volume data illustrates the trend: unit shipments for ADAS technologies are expected to reach about 334 million in 2024 alone, driven by demand for features like forward collision warning, automatic emergency braking, and adaptive cruise. As more of the fleet benefits from these functions, regulators and safety programs increasingly judge performance not against a human baseline but against what electronic perception should reliably deliver.
Yet, a gap remains. Most current systems still assume a human driver remains legally and technically in charge. Level 4 changes this equation. Instead of supporting a driver, the system becomes responsible for maintaining safe behavior inside a defined operating domain, with the expectation that it manages rare weather, traffic, and infrastructure edge cases without relying on a human to intervene. That shift elevates redundancy, fail-operational behavior, and fleet operations from implementation details to core elements of the safety case.
Bhanot bridged this gap directly. Before joining Applied Intuition’s trucking program, he worked on ADAS validation at Ford Motor Company, gaining experience with large-scale simulation. He later transitioned to L4 trucking at Applied Intuition, helping scale a perception team from two to over twenty engineers. He led the design of two generations of multi-modality sensor suites for Level 4-style trucking operations, taking configurations from concept through public-road deployment with partners in Japan.
Those architectures enabled an initial fleet of 3 trucks operating across the United States and Japan, generating over 90,000 miles of real-world data to exercise handover logic, failure handling, and edge-case perception in the field rather than treating them as theoretical scenarios. The second-generation Bhanot worked on is expected to enable a fleet of 30 autonomous vehicles by the end of fiscal year 2027. “The challenge isn’t just detecting an object,” says Bhanot. “Level 4 assumes the system, not the driver, handles the hard scenarios. In Japan’s GPS-sparse tunnels or on US highways with sometimes faded lane markers, the system cannot ask for help; it must have a pre-computed safe state.”
Designing Redundancy For Level Four Platforms
Once automation is treated as the primary operator, redundancy becomes a foundational requirement rather than a safety add-on. Freight and logistics use cases show why. The global autonomous trucks market is estimated at around $1.74 billion in 2025, with forecasts pointing to strong growth through the next decade as long-haul corridors and hub-to-hub routes adopt higher levels of autonomy. That scale magnifies the consequences of a single-point failure, whether it is a sensor outage, network glitch, or computer fault, and it raises the bar for what counts as an acceptable degraded state.
In Level 4 contexts, redundancy is not just about having extra hardware; it is about ensuring that multiple sensing and compute paths can independently support safe behavior when something goes wrong. “Redundancy is not just having two cameras. It is about modal independence,” says Bhanot. This requires mapping failure modes to concrete design decisions: how fields of view overlap, how power domains are separated, how communication backbones are isolated, and how perception and planning algorithms are structured so that one failing component does not silently erode risk margins. The goal is to maintain safe operation or controlled exits even when the system is operating with partial capability.
At Applied Intuition, Bhanot made that mapping explicit. Leading the perception systems engineering function for the trucking division, he translated sensor failure modes, occlusion patterns, and environmental limits into specific layout, modality, and compute choices within the platform architecture. Redundant coverage was designed not only for nominal performance but also for degraded modes, with overlapping fields of view, disciplined power and network topologies, and clear assumptions about which subsystems had to remain available for the vehicle to continue operating. By encoding those patterns into architecture standards, he ensured that new routes, features, and configurations all aligned with a system-level redundancy strategy rather than accruing as isolated exceptions. “Redundancy means no single fault can quietly weaken safety,” notes Bhanot.
New Operating Playbooks For L4 Fleets
As technology teams harden redundancy, operations leaders are discovering that Level 4 programs require new playbooks to address the crippling cost of downtime. Just as a 2024 analysis found that industrial outages can cost up to $2.3 million per hour, similar economics apply to L4 fleets, where downtime is estimated to cost between 448 to 760 USD per day per vehicle. At scale, where automated fleets may be tightly integrated into supply chains, these economics mean that every incident is an operational stress test.
For companies, that reality shifts autonomy from a pilot on the side of the business to a core asset that has to be planned, monitored, and supported with the same discipline as any other critical infrastructure. Dispatch decisions, maintenance cycles, on-call readiness, and data feedback loops become as strategic as the onboard software stack. The organizations that succeed are likely to be those that treat L4 operations as a joint responsibility across engineering, safety, and fleet management from the outset.
Bhanot’s work on the Applied Intuition trucking program treated this as a measurable engineering problem rather than an abstract aspiration. He defined validation metrics and test strategies that explicitly targeted failure handling, built around scenario libraries that could be reused as the architecture evolved. Beyond the code, he helped set hiring, onboarding, and operating rhythms for the systems function, ensuring that every new route and regression test fed back into a shared understanding of the system’s limits. This rigorous approach extends to the research community, where Bhanot serves as an invited paper reviewer for the IEEE Internet of Things Journal and the IEEE Transactions on Intelligent Transportation Systems (T-ITS) Journal. In these roles, he connects day-to-day fleet realities with the standards emerging in academic research. “A Level 4 operation must prove behavior under stress, not just when conditions are ideal,” states Bhanot, adding that “Level 4 only scales when engineering, safety, and fleet ops share the same playbook.”
Looking Ahead, Building Trustworthy L4 Systems At Global Scale
As freight, passenger, and mixed-use programs move from pilots to early commercial deployments, long-term projections are starting to look more like planning baselines than speculation. One forecast expects the autonomous vehicles market to reach about $83.1 billion by 2035, underscoring how deeply autonomy is likely to be woven into global mobility and logistics networks over the next decade. That trajectory raises an important question for industry and regulators alike, who will define the norms for redundancy, fail-operational behavior, and fleet operations as these systems scale.
For Bhanot, that responsibility sits at the intersection of engineering practice and governance. His work architecting Level 4 sensing and perception systems for global trucking fleets, combined with his editorial and judging roles in technical communities, positions him among the engineers quietly setting expectations for what trustworthy autonomy looks like in the field. As L4 deployments accelerate, voices that can connect architecture, safety cases, and operating playbooks will play an outsized role in determining whether autonomy is experienced as a dependable utility or a fragile experiment. “Level 4 earns its place when it feels predictably boring because it is consistently safe,” says Bhanot.
