AD stacks versus SDV.OS stacks

I ask myself what are the mayor differences between the todays AD stacks such as NVIDIA Drive (OS, Sim, Works) or BAIDU Apollo (Auto) and the Software Defined Vehicle (SDV) operating system (OS) approaches.

OS & middleware components

In a first sense it looks quite similar:

• Operating System (OS)
  • Board Support Packages (BSPs)
  • Over the Air Update (OTA)
  • Hypervisor (on OS, process or activity level)
• Development environment
  • Simulation environment
  • Build environment
  • SDK (tooling, API, supported programing languages)
• Middleware
  • Communication focus e.g., DDS, IPC
  • other functional management areas e.g.,
    • Execution management such as program states (started, stopped, etc.)
    • Log management such as how to store log data

Often with a more or less explicit control plane.

Data plane aspects

In my opinion the mayor difference is with the data plane and the data processing.

• Sensor and actuator abstraction e.g., via respectiv API
  • Environment sensing e.g.,
    • LIDAR
    • Radar
    • Camera
  • Positioning e.g., GPS
  • Speed e.g., IMU
  • Bus systems e.g.,
    • CAN
    • LIN
  • Time and timing
  • Communication (ethernet, IPC, Inter SoC)
• Data management & Algorithm keeping the control loop alive. Most distinguishing part is
  • the processed data volume and the link back to the simulation (record and replay) plus the aspects mentioned on https://www.teachkidsrobotics.com/what-is-a-robot/
    • the supported consistent: The robot is able to perform the same task repeatedly the same way (the action is deterministic)
    • the supported accurate: The robot can perform its actions with a high degree of accuracy
    • the supported reliable: The robot should not fail/require maintenance often.

Control loop components

Control loop (Processing, Detection, Perception equal to robotics Sense, Think/Plan, Act)

• Data processing e.g.,
  • Point cloud processing
  • Image/ video processing
• Information detection out of processing e.g.,
  • Wait condition e.g., for traffic light
  • Object detection
• Semantic perception out of detection e.g.,
  • Landmarks = obstecal
  • Lanes
  • Safe force
  • Traffic light
  • Free space
  • Traffic sign

For every perception component a respective set of tasks is to over e.g., for lanes a respectiv lane keeping.