The dynamic window approach is a fast local planner that picks a robot's next velocity by simulating short candidate motions and scoring them — the reactive layer that follows a global plan while dodging obstacles in real time.
The dynamic window approach decides how a robot should move in the next moment. It imagines several possible speeds and turns, predicts where each would take it, and picks the one that heads toward the goal while avoiding obstacles — recomputed constantly as it drives.
A global planner draws the overall route, but a robot still needs to decide, moment to moment, exactly how fast to drive and turn while avoiding obstacles that appear. The dynamic window approach (DWA) is a classic, fast way to make that call.
How it works
DWA plans in velocity space, not position. Each control cycle it:
Builds the dynamic window — the set of (linear velocity, angular velocity) commands the robot can actually reach in the next instant, given its current speed and acceleration limits (hence "dynamic": it respects what the robot can physically do right now).
Simulates each candidate velocity forward a short time, predicting the short arc the robot would follow.
Scores each trajectory by a weighted objective: progress toward the goal, clearance from obstacles, and forward speed. Trajectories that would hit something are discarded.
Executes the best command — then repeats next cycle.
Simulate short moves, pick the best
Every cycle DWA evaluates the motions the robot can immediately make and picks the one that best balances heading to the goal and avoiding obstacles.
Why it's useful
Fast and reactive. Cheap enough to run at high rate, so the robot dodges dynamic obstacles smoothly in real time.
Respects real limits. By planning in reachable velocity space, it never commands accelerations the robot can't achieve — its motions are always feasible.
Pairs with a global planner. DWA is a local planner: it follows the global path (A*/D*) while handling immediate obstacles from the live costmap. This global-plus-local split is standard in Nav2.
The limits
Local minima. Like other purely local/reactive methods (compare potential fields), DWA can get stuck when short-horizon scoring can't see a way out — which is why it relies on a global planner and recovery behaviors.
Short horizon. It only looks a short time ahead, so it's about immediate reaction, not global optimality.
Successors like the Timed Elastic Band optimize a longer local trajectory for smoother, more anticipatory motion.
Why it matters
DWA is a foundational local navigation method — the reactive velocity planner that lets a robot follow its route while safely handling the obstacles it encounters, all within its physical limits. It's a default building block of mobile-robot navigation stacks.
