Motion planning is the part of an autonomous system that decides where to go and how to get there. Perception tells the vehicle what is around it; control turns commands into wheel and steering actions; motion planning is the layer in between that produces the actual path — the route through space, and often through time, that the vehicle will follow to reach its goal without hitting anything. The field uses three overlapping terms. Path planning usually means the geometric route: the sequence of positions. Trajectory planning adds the time dimension: where the vehicle should be at each instant, and therefore its speed and acceleration along the route. Motion planning is the umbrella that covers both. In the patent record these inventions classify principally under CPC G05D, "Control of position, course, altitude or attitude of land, water, air or space vehicles," and under the conjoint vehicle-control subclass B60W.

A motion-planner claim is recognizable by its shape: it perceives a situation, generates one or more candidate paths, selects a path under some criterion, and commands the actuators to follow it. US9229453B1, "Unified motion planner for autonomous driving vehicle in avoiding the moving obstacle," a granted patent assigned to GM Global Technology Operations LLC, lays the structure out step by step in its claim 1.

detecting a lane center of a roadway lane that the vehicle is traveling along; determining a lane centering path that directs the vehicle from a current vehicle position to the lane center; automatically steering the vehicle along the lane centering path; detecting a moving object in front of the vehicle— US9229453B1, claim 1, source

The claim then walks the decision tree a planner has to evaluate. After detecting a moving object, it recites "determining if a collision between the vehicle and the object will occur if the vehicle travels along the lane centering path at a current vehicle speed," then "determining whether vehicle braking alone will prevent the collision," then "determining a steering path around the object if it is determined that vehicle braking will not prevent a collision," and finally "steering the vehicle along the steering path." Notice the logic: the planner has a default plan (stay centered in the lane), tests whether that plan collides, tries the cheapest evasion first (brake), and only computes a steering-around path if braking is insufficient. That escalation — default path, predicted collision, braking check, evasive path — is the substance of what the claim fences off. It is not "motion planning" in the abstract; it is a specific obstacle-avoidance decision sequence with braking preferred over steering.

What a planner claim actually locks down

The value of reading the independent claim of a motion-planner patent is that planning is an enormous, decades-old field — A*, RRT, lattice planners, model-predictive control, and optimization-based trajectory generation are all extensively published and largely public. A modern planner patent therefore cannot own "planning a path"; it owns the particular decision structure it recites. In US9229453B1 the narrowing substance is the lane-centering default plus the brake-before-steer escalation. A different planner patent might claim a particular candidate-generation method, a specific cost function for ranking paths, a re-planning trigger, or a constraint on how aggressively the vehicle may swerve. Two patents both classified for autonomous motion planning can fence off entirely non-overlapping decision logic. The class G05D tells you the record is about vehicle course control; only the claim tells you which planning decisions it covers.

Path versus trajectory in the claim language

The path-versus-trajectory distinction often shows up directly in claim limitations and is worth tracking. A claim that recites only positions — "a steering path," "a set of spatial points," "target touchdown locations" — is a path-planning claim; it constrains geometry. A claim that adds speed, time, or state-at-time limitations — "a desired vehicle state for a time after a first update interval," "a capture point trajectory," "velocity" — is a trajectory-planning claim; it constrains motion through time. The difference is legally meaningful: a competitor that plans the same geometric route but with a different speed profile may fall outside a trajectory claim while infringing a pure path claim, or vice versa. When mapping a planning portfolio, separating the geometric-path claims from the timed-trajectory claims reveals which dimension of the planning problem a company has actually fenced off.

There is also a prediction dimension hiding inside many planner claims that is easy to miss. US9229453B1 must reason about "a moving object in front of the vehicle" and "determining if a collision between the vehicle and the object will occur if the vehicle travels along the lane centering path at a current vehicle speed." To make that determination, the system has to forecast where the object will be over the planning horizon, not just where it is now. Planner claims that recite collision prediction therefore embed an implicit motion model of other agents, and some patents narrow specifically on how that prediction is performed. When two planner patents both claim obstacle avoidance, the one that fences off a particular prediction-and-avoidance coupling covers different ground than one that assumes a static or simply-extrapolated obstacle. Reading the claim for whether — and how — it predicts the obstacle's future position is part of sizing what it actually protects.

A practical search note closes the picture. Because planning straddles perception and control, motion-planner patents commonly carry both a G05D course-control symbol and a B60W conjoint-control symbol, and sometimes a perception or machine-learning class for the prediction component (predicting where the moving obstacle will be). A portfolio map seeded only on one class will undercount the estate. Seed on G05D and B60W together, then read each independent claim for its specific decision logic — default plan, collision test, evasion strategy — because that logic, not the class symbol or the title, is what the patent fences off.