Skip to content

Vehicle Dynamics and Control Fundamentals

Vehicle Dynamics and Control Fundamentals curated visual

Visual: model-fidelity ladder from kinematic bicycle to dynamic bicycle, tire force/slip, actuator delay, controller output, and failure diagnostics.

Planning decides where the vehicle should go. Control makes that motion happen through tires, motors, brakes, steering actuators, hydraulics, and software limits. The planner-controller boundary is where latency, saturation, slip, comfort, and safety constraints become real.

This page covers the vehicle models and control methods needed to turn trajectories into executable commands for road AVs, indoor AMRs, outdoor industrial vehicles, and airport ground vehicles.


1. AV, Indoor, Outdoor, and Airside Relevance

DomainDynamics/control concernExample
Road AVHigher speeds, tire slip, lane tracking, comfort, actuator delay, fail-operational steering/braking.MPC tracks a 10 Hz trajectory while compensating steering delay and curvature limits.
Indoor AMR / forkliftLow-speed nonholonomic motion, reversing, payload changes, tight docking, safety-rated stops.A forklift with load changes its braking distance and rear-swing envelope.
Outdoor yard / mine / campusUneven ground, tire-soil interaction, dust/rain, heavy vehicles, grade, limited friction.A yard tractor slips on wet painted markings and must reduce speed and curvature.
Airside AVLow-speed precision near aircraft, wet tarmac, jet-blast zones, stand entry, mixed GSE traffic, strict passenger/crew comfort.A baggage tractor must track a stand approach path smoothly while preserving emergency-stop margin.

2. Model Hierarchy

ModelState detailBest useLimit
Point massposition, velocity, accelerationSpeed planning, rough feasibility, comfort envelopes.No heading, steering, or nonholonomic constraint.
Unicyclex, y, yaw, vSimple mobile robots and early planning.Does not represent Ackermann steering geometry.
Kinematic bicyclerear/front axle pose, steering angle, speedLow-to-moderate-speed path tracking and trajectory feasibility.Ignores tire slip and load transfer.
Dynamic bicyclelateral velocity, yaw rate, tire forcesHigher-speed control and friction-limited maneuvers.Needs tire parameters and accurate velocity/slip estimation.
Double-track / multibodyfour wheels, suspension, load transferHigh-fidelity simulation, validation, extreme maneuvers.Too complex for most real-time control loops.
Learned residual modelmodel error correctionCompensates delay, tire/terrain effects, payload.Must be bounded and monitored for safety.

Use the simplest model that captures the dominant risk at the operating speed and surface. Airside baggage tractors and indoor AMRs are often low speed, but payload, wet surfaces, reversing, and docking precision still make model selection important.


3. Kinematic Bicycle Model

For an Ackermann-steered vehicle with wheelbase L, speed v, steering angle delta, and heading theta:

x_dot     = v * cos(theta)
y_dot     = v * sin(theta)
theta_dot = v * tan(delta) / L
v_dot     = a

Curvature is:

kappa = tan(delta) / L
delta = atan(L * kappa)

Physical steering limits become curvature limits:

|delta| <= delta_max
|kappa| <= tan(delta_max) / L

Steering actuator limits also impose curvature-rate limits:

|delta_dot| <= delta_dot_max

This model is usually sufficient for:

  • low-speed airside and warehouse vehicles
  • Frenet trajectory feasibility checks
  • pure pursuit, Stanley, LQR, and many MPC controllers
  • simulation smoke tests

It becomes weak when side slip, high lateral acceleration, rough terrain, or heavy load transfer dominate.


4. Dynamic Bicycle Model

The dynamic bicycle model adds lateral velocity and yaw-rate dynamics. A common state is:

[x, y, yaw, v_x, v_y, r]

where:

  • v_x: longitudinal velocity in body frame
  • v_y: lateral velocity in body frame
  • r: yaw rate

Approximate tire slip angles:

alpha_f = delta - atan((v_y + l_f * r) / v_x)
alpha_r =       - atan((v_y - l_r * r) / v_x)

Linear tire model:

F_yf = C_f * alpha_f
F_yr = C_r * alpha_r

Yaw and lateral dynamics:

m * (v_y_dot + v_x * r) = F_yf + F_yr
I_z * r_dot             = l_f * F_yf - l_r * F_yr

The dynamic model exposes risks hidden by the kinematic model:

  • tire saturation
  • sideslip
  • understeer/oversteer
  • payload and center-of-gravity changes
  • friction-circle limits

At low speeds, the equations can become numerically sensitive because v_x appears in denominators. Controllers need low-speed handling rather than blindly applying high-speed dynamic equations near zero velocity.


5. Actuator and Interface Realities

5.1 Command Types

CommandTypical interfaceHidden complexity
steering angleradians or steering wheel anglesteering ratio, offset, slew limit, backlash, calibration.
steering rateradians/sactuator delay and saturation.
accelerationm/s^2 targetmaps to throttle/brake differently by speed, slope, payload.
throttle/brakenormalized or torque requestactuator nonlinearity, deadband, regen blending.
gear/directionforward, reverse, neutralplanner must understand nonholonomic reverse motion.
emergency stopsafety input or brake commandstopping distance and jerk must be modeled separately from nominal control.

5.2 Delay

A simple delay model:

u_applied(t) = u_commanded(t - tau)

Delays come from perception, planning, middleware, control, DBW, hydraulics, brake pressure build-up, and steering mechanics. A controller that ignores delay will oscillate or cut corners as speed rises.

Practical mitigations:

  • timestamp all trajectories and states
  • predict vehicle state to actuation time
  • include delay in MPC state or command buffer
  • identify delay from step-response tests
  • set speed limits based on worst-case latency

6. Controller Families

ControllerUseStrengthRisk
PIDlongitudinal speed, simple steering loopsEasy to implement and tune.Windup and poor constraint handling.
Pure pursuitgeometric path trackingRobust and simple at low speed.Cuts corners; lookahead tuning is speed-dependent.
Stanleylateral path trackingHandles cross-track and heading error with speed scaling.Needs low-speed softening and steering saturation handling.
LQRlinearized tracking around trajectoryGood stability and efficient compute.Local validity depends on model and linearization.
MPCconstrained multivariable trackingHandles curvature, acceleration, jerk, delay, and actuator limits.Requires model quality and solver timing guarantees.
iLQR / DDPnonlinear trajectory optimization/controlEfficient for smooth nonlinear systems.Constraint handling and warm starts need care.
MPPIsampling-based stochastic controlHandles nonlinear costs and rough models.Compute and stochastic variability.
Safety filter / CBF-QPlast-line constraint enforcementEnforces safety set over nominal commands.Feasibility and false interventions must be validated.

6.1 Lateral and Longitudinal Coupling

Even if a stack has separate lateral and longitudinal controllers, the vehicle does not. Curvature limits speed through lateral acceleration:

a_lat = v^2 * |kappa|
v_max = sqrt(a_lat_max / |kappa|)

The speed planner must slow down before high curvature, poor friction, docking, crowds, or uncertain localization. Otherwise the lateral controller is asked to track an infeasible path.


7. Practical Deployment Notes

7.1 Identification Checklist

Measure or estimate:

  • wheelbase and track width
  • vehicle reference point
  • steering ratio, offset, maximum angle, and maximum rate
  • throttle/brake delay and deadband
  • maximum acceleration, deceleration, and jerk
  • tire/friction assumptions for dry, wet, icy, painted, dusty, or oily surfaces
  • loaded vs unloaded mass and center of gravity
  • reverse-driving behavior
  • controller compute and communication latency

7.2 Controller Interface Contract

A trajectory handed to control should include:

  • timestamp and frame
  • pose, velocity, acceleration, curvature, and optionally curvature rate
  • allowed speed and acceleration envelopes
  • stop point and emergency fallback point
  • validity horizon
  • covariance or confidence where available

The controller should report:

  • tracking error
  • command saturation
  • delay estimate
  • actuation fault state
  • degraded-mode reason
  • predicted stopping distance

7.3 Airside and Industrial Policy

For airport and industrial vehicles, conservative control policy is usually more valuable than high dynamic performance:

  • cap speed near aircraft, personnel, open holds, baggage carts, and stands
  • reduce curvature and jerk on wet or contaminated pavement
  • preserve safe-stop energy and braking margin
  • use separate docking controllers for final centimeters
  • require manual/service checks after actuator saturation or repeated tracking faults

8. Failure Modes and Risks

Failure modeSymptomMitigation
Planner outputs infeasible curvatureController saturates steering and cuts path.Check curvature, curvature rate, and minimum turning radius in planning.
Actuator delay ignoredOscillation, overshoot, or corner cutting.Identify delay and compensate in state prediction or MPC.
Tire slip / low frictionVehicle understeers, oversteers, or braking distance grows.Estimate friction, reduce speed, monitor yaw-rate residuals.
Integral windupLongitudinal controller overshoots after saturation.Anti-windup, saturation-aware PID, or MPC.
Low-speed singularityDynamic or Stanley-style terms divide by near-zero speed.Add low-speed modes, softening terms, or geometric docking controllers.
Payload changeBraking and steering response change.Include load state or conservative envelopes.
Reverse/crab mode mismatchController assumes forward Ackermann motion.Explicit mode in state and trajectory contract.
Bad localization velocityController chases noisy state estimates.Filter velocity, validate timestamping, and gate degraded state.
Solver overrunMPC command arrives late or stale.WCET monitoring, fallback controller, bounded iterations.
Safety filter infeasibleLast-line controller cannot find a safe command.Design reachable fallback states and stop before constraints become impossible.


Sources

Public research notes collected from public sources.