drivenext.ai

planning, UX

How a Self-Driving Car Chooses Its Next Path

Planning is not finding a route across town. It is picking the next few seconds out of a huge set of options.

EducationalNot safety advice

In a strategy game, most moves are legal and almost all of them are bad. Skill is not generating options. It is scoring them quickly and throwing the losers away.

Driving planners work the same way. Given where things are and where they are likely going, the system considers many short-horizon trajectories: hold the lane, ease left, lose two miles per hour, wait one more second before turning. Each option is scored against several goals at once — avoid collisions, follow rules, make progress, stay comfortable, stay readable to other people.

Skill is not generating options.

Those goals conflict. The safest plan is to never move. The fastest plan is uncomfortable or unsafe. The most literal reading of a rule can still confuse everyone nearby, which is its own problem. Planning is a negotiation among goals that cannot all be maximized.

This is also where a system's personality comes from. Two stacks with similar perception can feel different to ride because one weights comfort higher and another weights progress. Hesitant, abrupt, serene — those labels are often planner priorities, not a difference in what the cameras captured.

And the plan is disposable. It is recomputed many times per second. Most candidate paths are discarded before they are ever sent to the actuators.

This site runs on drivenext.ai — the domain is for sale (Buy It Now / make offer on Afternic). Shu Legacy Operating LLC also lists related domains for builders: blockpulse.ai, cryptoforge.ai, and drivenext.co (bundle-only with drivenext.ai).
Afternic listing →