Traditional robotic navigation often treats path planning and velocity control as two separate problems. A high-level planner generates a geometric path, and a low-level controller attempts to track that path at a constant or loosely regulated speed. However, when a robot operates near its physical limits—such as an autonomous car taking a sharp turn, a drone navigating a tight tunnel, or an industrial arm moving a fragile object—this decoupled approach fails.
If a vehicle enters a sharp curve too quickly, it saturates its actuators or exceeds environmental friction limits, resulting in corner-cutting, spin-outs, or mechanical failure. Dynamic velocity scaling solves this by directly coupling a robot's longitudinal speed to the local geometry of its path and the physical constraints of its environment. By treating velocity as a dynamic variable rather than a fixed target, robots can safely push to the absolute edge of their performance envelopes.
The Geometry-Velocity Coupling
To understand why velocity must be coupled to geometry, consider the basic physics of cornering. The centripetal acceleration required to stay on a curve is given by $a_c = v^2 / R$, where $v$ is velocity and $R$ is the radius of curvature. If a robot attempts to maintain a uniform velocity as $R$ decreases (i.e., the turn becomes sharper), the required lateral force spikes quadratically. Once this force exceeds the available friction between the robot and the surface, control is lost.
As noted by Huang et al., traditional geometric steering methods that rely on fixed look-ahead distances inevitably suffer from tracking errors; they either oscillate at high speeds or cut corners on sharp curves. To prevent this, Bibuli et al. introduced heuristic speed adaptation rules that automatically reduce forward velocity when approaching curves. This creates an explicit coupling between yaw (steering) and surge (forward speed) control, leading to significantly smoother transitions and tighter tracking.
In highly agile environments like robot soccer, Agarwalla et al. utilize curvature-aware velocity profiling to calculate the maximum safe speed at every discrete point along a path. This optimization ensures that the robot balances traversal time with stability, operating right at the boundary of its friction and balance limits. Similarly, Polack et al. demonstrate that a local velocity planner computing target speeds based on upcoming road geometry guarantees that the planned path remains physically feasible before the low-level controller even attempts to execute it.
Velocity-Dependent Constraint Modeling
The physical limits of a system are rarely static; they change based on how fast the system is already moving. For example, a car cannot simultaneously exert maximum braking force and maximum cornering force—a concept visualized by the classic "friction ellipse."
To capture this interplay, Polack et al. model vehicle acceleration and cornering limits as velocity-dependent convex constraints. By fitting polynomials to the vehicle's acceleration envelope relative to its longitudinal speed, their planner dynamically adjusts the available grip margin. If the vehicle is traveling fast, the planner automatically restricts sharp steering commands because it knows the available lateral grip is diminished.
High Velocity Low Velocity
Friction Ellipse Friction Ellipse
+---+ +-------+
| | | |
-Ax ----+---+---- +Ax -Ax ----+-------+---- +Ax
| | | |
+---+ +-------+
-Ay (Limited Lateral Grip) -Ay (Maximum Lateral Grip)
This velocity-dependent framing is critical for complex, multi-body systems. For autonomous semi-trailer trucks, Wang et al. implement speed-dependent articulation constraints. At low speeds, the controller permits wide articulation angles to allow for agile maneuvering in tight spaces; at high speeds, it enforces strict, soft limits to prevent dangerous jackknifing.
The same principle applies to industrial automation. Zhang et al. map the friction and balance limits of unattached, fragile objects into a robot arm's dynamic constraints. This allows the manipulator to maximize acceleration on straight paths and decelerate precisely before curves, cutting handling times in half without dropping the payload.
Adapting to Environmental Uncertainty
In real-world deployments, a robot's physical limits shift dynamically due to changing environmental conditions. A velocity profile optimized for dry asphalt will cause a vehicle to slide out on ice or wet grass.
To maintain stability, modern architectures estimate environmental parameters online and update their velocity profiles in real time. Dolan et al. utilize an online learning framework to estimate the current tire friction coefficient from vehicle dynamics, instantly updating the global reference speed plan to prevent control saturation. To bypass the computational cost of re-optimizing trajectories from scratch on the fly, AL-Sunni et al. combine real-time friction estimation with a pre-computed library of velocity profiles. When track conditions degrade, the controller instantly swaps to a safer profile from the library, keeping the vehicle at its optimal safe speed.
For proactive safety, Kim et al. optimize longitudinal speed based on estimated cornering stiffness, enforcing a non-linear reduction in maximum velocity as road friction drops. Velocity modulation is equally critical when navigating rough terrain. As Manderson et al. demonstrate, integrating terrain roughness into velocity planning allows off-road vehicles to automatically slow down on high-vibration surfaces, mitigating mechanical stress and preventing loss of traction.
Generalizing Beyond Ground Vehicles
The intuition behind dynamic velocity scaling extends across diverse robotic domains, proving that speed modulation is a universal tool for managing complex physics.
- Aerial Manipulation: Guo et al. include execution time and contact force as decision variables in trajectory optimization for drones performing contact-rich tasks (like calligraphy). This forces the drone to automatically slow down during high-force or high-curvature strokes to stay within its actuator torque limits.
- Perception and Aerodynamics: When quadrotors fly through narrow tunnels, they encounter severe aerodynamic disturbances and visual tracking degradation. Wang et al. design a motion planner that slows the drone down during sharp maneuvers to preserve visual feature tracking, while accelerating through straight sections to escape unstable, low-speed turbulence zones.
- Marine Control: Unmanned surface vehicles are highly underactuated and prone to drifting. Feng et al. show that by adaptively reducing velocity during path deviations, a boat can execute much tighter turns, allowing for rapid course correction and dynamic obstacle avoidance with minimal loss of forward progress.
- Legged Locomotion: In simulated humanoids navigating complex 3D environments, Hallak et al. demonstrate that adaptive speed control based on terrain features (like stairs or steep inclines) prevents the catastrophic balance failures that occur when agents attempt to maintain a uniform walking velocity.
Ultimately, whether a robot is a race car exploiting tire slip angles to drift through a corner—as explored by Trumpp et al.—or an off-road rover executing a two-pass speed plan over rugged terrain, as shown by Saripalli, the core mechanism remains identical. By treating speed as a dynamic function of geometry, capability, and environment, autonomous systems achieve a superior trade-off between execution speed and operational safety.
Key Foundational Work
Path-following algorithms and experiments for an unmanned surface vehicle (2009) — contributed 2 concept elements to this direction.
Lateral control for Autonomous land Vehicles via Dual Heuristic Programming (2016) — contributed 1 concept elements to this direction.