Why it matters
Faults will happen. Safety is about detecting them in time and reacting so that no hazard results.
What the building blocks are
- Monitoring: watchdogs (is the software alive and on time?), plausibility checks, voltage and clock monitors, lockstep comparison.
- Fault tolerance time interval: how long a fault may persist before it could cause harm — the reaction must be faster.
- Safe state: e.g. motor torque off, valve closed, warning to the driver.
- Degraded / fail-operational mode: for functions that cannot simply switch off (e.g. steering at speed), a reduced but safe function continues on a redundant path.
Best practices
- Use an independent shutdown path that does not rely on the faulty software.
- Test fault reactions with fault injection on real hardware.
Common pitfalls
- A safe state that is not actually safe in all driving situations.