Rove

"Everyone thinks getting somewhere means heading straight at it," he cooed. "A naive robot does exactly that — the 'greedy' method: point at the goal, drive forward. Watch." The robot aimed straight and rammed a block, wheels spinning. "It wanted the goal. It just had no path. Wanting isn't a plan — and greedy-toward-the-goal gets stuck at the first wall."

He spread a grid over the maze. "Real navigation has two halves. First, the robot needs a map of what's in the way — an occupancy grid, where each little square is marked free or blocked from its sensors. You can't plan a route through a world you haven't mapped. Second, path planning: a search over that grid for a route of free squares from here to the goal. A good planner — the kind that powers real robots — explores outward, always expanding the most promising square first, until it finds the shortest clear path. And crucially, it will happily route away from the goal to get around an obstacle, because the shortest straight line is usually blocked and the shortest real path curves."

"Why can't it just turn when it hits the wall?" a builder asked.

"That's the greedy trap — the local minimum," Rove said. "Drive toward the goal, hit a wall, follow the wall... and a U-shaped obstacle can catch you in a pocket forever, endlessly 'closest' to the goal but never reaching it. A planner that looks at the whole map escapes the pocket, because it can see that the way out leads backward first. That's the difference between reacting and planning."

He traced a winding line. "Once the planner finds the route, it doesn't need every square — it picks waypoints, a few key turning points, and drives from one to the next, re-checking for anything new that moved into the way. Map the obstacles, plan a route over the map, follow the waypoints, replan if the world changes."

"So reaching a goal isn't about going straight at it," the builder said slowly. "It's about finding the path around what's in the way."