If you picture an autonomous vehicle, you probably imagine it seeing — reading the terrain, spotting obstacles, finding a way through. You probably think about colourful perception overlays that highlight drivable regions, obstacles and terrain features, giving the impression that autonomous navigation is largely solved. But seeing is the easy half. The real work begins the moment the vehicle has to choose.

Last week in From Data to Dirt, we described one example of how our perception algorithms learn to tell drivable ground from everything else. What comes out of that is a drivable area map, where movement is possible. On a paved road, that map almost answers the question by itself. Off-road, it doesn't — because "possible" and "sensible" are two very different things.

Cross a field or a forest track and you'll usually find not one path, but several. One is shorter but muddy. Another offers better traction but leads over a steep slope. A third is not ideal for traction or slope but seems a good compromise. So the question is no longer "Can I drive here?" but "Which of these should I take — and does it get me where the mission needs me to be?"

Answering that is what trajectory planning does. And it always starts from intent: the mission goal and the waypoints the operator has set. A simple logistics dropoff might change the center of gravity and, therefore, the slopes a vehicle can handle on the way back. The mission, together with the defined destination, the desired waypoints and the vehicle's current position, turn an open map into a real task — get from here to there, through this terrain and consider these constraints.

Within that task, the planner juggles competing demands all at once — slope, traction, obstacles, and the simple question of whether the vehicle can physically manage the move. Rarely do they point the same way: the shortest line isn't the safest, the safest isn't the fastest. The planner keeps weighing these against one another and settles on a trajectory that is not just possible on paper, but works in the dirt — one the vehicle can then follow on its own.

That balancing act is exactly what off-road autonomy demands, because there is no pre-drawn line to follow. The vehicle has to read its surroundings, weigh the options against the mission, and commit. Perception tells it what the world looks like. Planning tells it how to move through that world toward a goal. Only together do they add up to autonomous navigation.

Because autonomy isn't really about seeing. It's about deciding.