Classes / Fleet & Robotics class / Rover Ops Department
Rover Ops Track: Autonomy, Field Maintenance and Drones · Module 1 of 4 · about 30 min
How autonomous machines drive
SAE J3016 describes six levels of road-driving automation, from zero to five, based on the driving task and human role. Levels zero through two retain an engaged human driver. Level three includes a user available for requested takeover, while higher levels have different system responsibilities and scope. This road taxonomy is not a universal permission or capability rating for every mine robot, aircraft or terrain.
A useful functional model distinguishes perception, localization, planning and control. Perception interprets sensor observations; localization estimates position using available evidence and uncertainty; planning selects intended motion; control attempts to execute it. Actual architectures differ, and no one favorable function result establishes all the others.
Each sensor has weaknesses. Cameras struggle in glare and darkness, lidar can be affected by dust and heavy rain, and GNSS drops in pits and under cover. Systems combine sensors so one can cover for another. Dust and mud are why field techs clean and check sensors constantly.
An operational design domain describes supported conditions such as route, speed, weather and environment. The actual system needs its approved behavior when those conditions are no longer supported. A generic instruction to take over or stop does not establish a safe response for every kind of machine.
When we review an automation claim, we ask who performs the driving task, who must remain available and which conditions the evidence covers. SAE's road-driving levels help with those human-role questions. They cannot establish that an unfamiliar mine route, weather condition or task lies within a particular machine's supported operating domain.
Perception, localization, planning and control answer different questions. Detecting an obstacle does not establish the machine's position accurately, and a planned path does not establish that the actuator followed it. Localization is an estimate with uncertainty, not an exact position guaranteed by a sensor label. Multiple sensors can share a common limitation, such as contamination or a missing reference.
Worked example, fictional review: a system detects a barrier but has uncertain position relative to the mapped route. The detection alone cannot establish that its planned clearance is valid. A separate demonstration in clear daylight does not establish performance in an unassessed dusty environment outside its supported conditions.
Practice, on paper: we trace invented observations through the four functions and the stated operating domain. Our self-check separates demonstrated capability, uncertain estimate and untested condition. We drive no vehicle, clean no live sensor and change no operating limit.
Sources
Module quiz
10 questions, 8 correct to pass. You'll see what you missed and why, and you can retake it.