An autonomous system perceives its surroundings, makes decisions and acts without a person steering: a self-driving vehicle, a warehouse robot, an uncrewed boat. Autonomy is not one algorithm but a stack, and the hard part is behaving safely in situations nobody listed in advance.
Key concepts
- The autonomy stack: perception, localisation, prediction, planning and control, each feeding the next.
- Sensor fusion: combining cameras, lidar, radar, inertial and satellite positioning to cover each other's weaknesses.
- Decision making: behaviour trees, state machines, planners and learned policies.
- Safety and fallback: monitors, redundancy and a minimal-risk state when something fails.
- Validation: simulation, closed-course tests and statistical evidence, because you cannot test every case.
Vocabulary
- Operational design domain: the conditions in which the system is meant to work.
- Levels of autonomy: a scale from driver assistance to no human at all.
- Edge case: a rare situation that the design did not anticipate.
- Human in the loop: a person who supervises or can take over.
What to study, in order
- Robotics basics, coordinate frames and probability for uncertain measurements.
- A simulator with a simple vehicle, and a controller that follows a path.
- Add perception: detect obstacles from a camera or range sensor in recorded data.
- Behaviour logic with a state machine or a behaviour tree, including failure states.
- Read how safety cases and standards argue that a system is safe enough.