Managing a team of three project managers working with a single customer, like Audi, sounds simple. It isn’t. In many ways, it is one of the most exposed positions a PM can have.
When the team is small, there is nowhere to hide. Every PM is visible. Every weakness is amplified. Every success or failure reflects directly on how the team is run. There are no buffers, no layers, and no place to dilute accountability.
At this stage, the work is hands-on by necessity. Details are reviewed. Customer meetings are attended. Coaching happens in real time. The goal is not scale yet, it is craftsmanship. You are building a shared language: how to escalate, how to report, how to say no without damaging trust.
The biggest challenge here is alignment. Three PMs can easily become three different voices to the same customer. That is unacceptable. Consistency matters more than creativity. The team must sound like one voice, even when individual styles differ.
This exposure becomes even sharper when the project itself is deeply technical and safety-critical, such as a mono-camera automotive vision system.
These programs are often described as software projects. In reality, they are full system programs, and things start to go wrong the moment that distinction is ignored.
A mono-camera project begins long before the first algorithm is integrated.
It starts at the hardware boundary.
Even when the silicon is predefined, for example a Mobileye EyeQ, the integration effort is substantial. The Tier-1 PCB design becomes a real concern, not because it is designed in-house, but because every later limitation can usually be traced back to early decisions made here.
Power delivery margins, signal integrity, memory interfaces, sensor connectivity, these are not abstract electrical details. They quietly define what performance is even possible. A PM team that treats this as someone else’s problem will eventually find itself explaining why targets were never realistically achievable.
Thermal considerations follow immediately and are often underestimated.
A vision system that performs well in the lab but throttles in real vehicle conditions is not suffering from a software defect. It is paying the price for early tradeoffs that were never challenged. Device placement, heat dissipation paths, enclosure assumptions, and ambient vehicle conditions must be addressed before designs are frozen. Once ignored, they resurface later as customer escalations that no schedule compression can fix.
Then comes communication.
Bandwidth, latency, bus selection, redundancy, and behavior under degraded conditions are system-level contracts. Camera-to-SoC links, SoC-to-vehicle networks, diagnostics channels, and OTA paths each dictate what data is available, when it is available, and with what guarantees. Weak early communication design leads to late-stage heroics and endless workarounds.
In parallel to the hardware efforts, the PM should also take care of the camera and the optics.
Lens selection, sensor choice, and camera positioning define the quality and limits of everything that follows. Field of view is always a tradeoff between coverage and resolution. A wider FOV may help detection range but degrades pixel density exactly where long-range performance is required. Color versus texture impacts night performance, robustness, and algorithm complexity. Resolution affects accuracy but drives bandwidth, memory, and compute pressure. Exposure control, HDR behavior, motion blur, and low-light noise directly shape algorithm stability in night, rain, glare, and tunnels.
Camera placement introduces its own constraints. Height, pitch, roll, windshield curvature, and vehicle-specific tolerances all affect perception geometry. Calibration accuracy and long-term stability are not optional details. They determine whether a system remains reliable over temperature changes, vibrations, aging, and service events. Optical distortion, focus drift, contamination, and sun load are not corner cases. They are everyday realities that must be designed for from day one.
Ignoring these optical and camera considerations early guarantees disappointment later. Algorithms cannot recover information that never reached the sensor.
Compute comes next, and compute is never infinite.
Every feature competes for cycles. Frame rate, resolution, neural network depth, redundancy, monitoring, and safety mechanisms all draw from the same budget. Someone has to force prioritization early and defend it consistently. The team must be aligned when explaining that adding one capability inevitably constrains another. Anything else is simply dishonest.
Only after these foundations are set do algorithms truly enter the picture.
Feature design and implementation, detection, classification, tracking, lane modeling, free space, all exist on top of physical and architectural assumptions. Required performance must be explicitly tied to system constraints such as lighting, vehicle speed, camera placement, compute headroom, and environmental variability.
And performance alone is never enough.
Automotive vision operates under certification regimes such as functional safety, SOTIF, ASPICE, and OEM-specific processes. These are not compliance checklists. They shape architecture, timelines, staffing, validation strategy, and change control. Ignoring them early guarantees friction, rework, and loss of credibility later.
This is where things become brutally honest.
Each PM must understand not only where the system excels, but also where it breaks. Each must know which problems can be mitigated later and which are fundamentally constrained by physics, silicon, or regulation. And all three must present a consistent narrative to the customer: what is fixed, what is flexible, and what is non-negotiable.
There is no room for mixed messages.
In a small team, the customer does not see individuals. They see a single system. Any inconsistency becomes a trust issue. Any optimism unsupported by system reality becomes technical debt, with interest.
This leads to an early but critical realization.
Your role is not to be the best project manager in the room.
Your role is to make sure the team does not depend on you to function.
The goal is not to absorb complexity for the team, but to teach systematic thinking, from PCB layout to perception algorithms, from thermal margins to certification gates.
When that happens, something changes.
The team stops waiting for answers. They start making aligned decisions.
And that is the moment you know you are no longer just managing projects. You are building leaders who can operate under total exposure.
#ADAS #AutomotiveEngineering #ComputerVision #SystemArchitecture #AutonomousDriving #ISO26262 #SOTIF #EmbeddedSystems #Tier1 #HardwareSoftwareIntegration
#ProgramManagement #ProjectManagement #TechnicalLeadership #TeamAlignment #Accountability #LeadPM #PMO #ProductLifecycle #StakeholderManagement
#LeadershipDevelopment #Mentorship #CareerGrowth #EngineeringManagement #SystemsThinking #ProfessionalDevelopment
#Mobileye #AutomotiveIndustry #TechStrategy #OEM #Innovation
Comments
Post a Comment