On city streets where autonomous shuttles roam, a safety valve lurks behind the scenes: a human hand that can guide or override the machine from afar. The new paper from Technische Universität München sketches a blueprint for how a control center could orchestrate a fleet of automated vehicles on public roads, stitching together safety, regulation, and human judgment in a single workflow. At the heart of the study is a simple but ambitious claim: if you map out every action a remote operator might take, you can build a universal sequence that keeps the wheels turning when automation falters.
Compiled by Maria-Magdalena Wolf and colleagues at TU Munich, with collaborators from EasyMile and industry partners, the work turns a murky domain—teleoperation for passenger shuttles—into something you can test, discuss, and legislate against. The team doesn’t pretend teleoperation is a silver bullet. Instead, they propose a concrete, testable structure: a control center framework that assigns tasks to two roles, Remote Operator and Fleet Manager, and a state diagram that lays out how an autonomous vehicle transitions from first connection to a safe, unmonitored drive again.
This is not a single technology, but an architecture for collaboration between people and machines. It’s about who does what and when, so that a fleet of self-driving vehicles can be supervised, supervised well, while still hugging the autonomy that makes them useful in the first place. The authors argue their framework should be validated on public roads and aligned with laws in different markets. That combination—technical clarity plus regulatory mindfulness—feels like a blueprint for responsible innovation rather than a rush toward full autonomy.
Roles and tasks in a teleoperation hub
The study distinguishes two human roles in a control center: the Fleet Manager, who oversees the health and scheduling of an entire fleet, and the Remote Operator, who works with one vehicle at a time, peering through screens and cameras instead of a windshield. The Fleet Manager watches for anomalies across dozens of vehicles, handles mission planning, charging schedules, and communication with infrastructure. The Remote Operator dives into the moment, translating what the vehicle reports and what the street looks like into concrete actions or guidance. This split mirrors how air traffic control separates tower duties from airline management; it’s a division of cognitive labor designed to prevent overload and keep the system reliable when the traffic gets dense or strange.
Because the paper draws on a broad literature base, the authors categorize driving-related tasks into three buckets: Remote Assistance, Remote Driving, and Remote Intervention. In Remote Assistance, operators offer high-level guidance, classify unknown objects, or propose maneuvers without directly controlling the car. In Remote Driving, the operator assumes real-time control, either steering and accelerating themselves or guiding the vehicle with low-level, direct input. Remote Intervention sits at the edge—when the vehicle needs a shove, a stop, or a reset—and can involve overriding the autonomous system to ensure safety. The framework also includes Remote Monitoring, a watchful phase where the operator tracks the vehicle and intervenes only if something looks off. This taxonomy isn’t just academic—it’s a practical map for how human and machine work together when the road becomes unpredictable.
The roadmap culminates in a state diagram that lays out the sequence from first connection to a safe, unmonitored drive. The diagram is deliberately generic so it can be adapted to different markets and laws. The TU Munich team emphasizes that the diagram covers the core driving-related tasks necessary to resolve disengagements—moments when the automation hesitates or misreads the world and needs a human hand to steer it back on course. In short, the framework answers a basic question: what does a remote operator actually do, and in what order, when an AV hits a snag on a public street? The answer, according to the authors, is a well-defined choreography rather than ad-hoc tinkering in the moment.
A universal blueprint born from literature
The team’s method begins with a long literature review, translating scattered, specialized descriptions into a single, coherent process. They pulled from prior work on control centers, remote input systems, and teleoperation concepts, then merged these strands into a generic task list and a process flow. The result is not a single company’s solution but a framework that other researchers and practitioners can adapt, test, and critique. The ambition is to provide a common language for how a remote operator and a fleet manager interact with an autonomous vehicle in a way that can be audited, regulated, and improved over time.
Key pieces of the architecture revolve around the ADS, the automated driving system, and its relationship to the vehicle’s state of operation. When the vehicle is in an MRC (minimal risk condition) or a safe stationary state, the remote operator can either approve continued automated driving, request an MRM (minimal risk maneuver), or switch the vehicle to a different mode of operation. The state diagram maps these possibilities, tracing the path from activation to a monitored automated state. In practical terms, that means the remote operator is not just pressing a button; they are interpreting sensor data, vehicle telemetry, and road geometry to decide the best next move within a disciplined, repeatable sequence.
The framework’s emphasis on transitions is what sets it apart. Instead of a loose set of guidelines, the diagram provides explicit state changes and the conditions that trigger them. This matters because disengagements—moments when the AV can no longer handle a scenario—are precisely where people must step in. By enumerating the states and transitions, the authors give engineers a blueprint for building teleoperation interfaces and rehearsing the decisions a remote operator would face, ensuring consistency and safety across a fleet rather than a patchwork of ad hoc responses.
Legislation, testing, and the road ahead
When the authors apply their diagram to German regulations, they don’t pretend the law mirrors the speculative elegance of a whiteboard diagram. German law currently prohibits Remote Driving in some contexts, so the state diagram must be adjusted to skip that transition. This is a sobering reminder that technology cannot outrun policy, and that a credible blueprint must bend to the real rules of the road. The authors demonstrate how the same architecture can accommodate different jurisdictions by swapping or removing interactions that aren’t allowed, without ripping up the entire diagram. It’s a pragmatic compromise that preserves the logic while staying compliant.
Beyond legality, the paper positions itself as a launchpad for real-world experimentation. The researchers envision testing the framework on public roads as part of larger pilot programs, with certification bodies verifying compliance and safety. They point to initiatives like the SUNRISE safety assurance framework as a possible guide for evaluation. In other words, this is not a theoretical exercise; it is a roadmap for how to prove, step by step, that remote support can meaningfully augment AV safety without turning the road into a testing ground for a new kind of human-in-the-loop chaos.
What makes the work feel timely is its balance of ambition and humility. The framework is a concrete, testable structure that can help operators handle a high variety of disengagements, while acknowledging that the exact user interfaces, teleoperation concepts, and response times still need careful design, testing, and training. The authors close by noting that future AVs might handle many MRCs themselves as software becomes smarter, but the current reality calls for a clear, auditable process for human support. The TU Munich team’s contribution—an organized, adaptable blueprint—might not make self-driving buses disappear tomorrow, but it could make them safer, more reliable, and easier to regulate as they move from pilot lanes to everyday streets.
