Open Source Robotics

A curated collection of high-quality open source robotics projects, tools, and software to propel the robotics community forward.

Frenet Frame Planning

  • Also called: Frenet planner, Frenet optimal trajectory planning, Frenet-frame trajectory generation
  • Category: Planning & manipulation › Motion planning
  • Tags: trajectory-generation, autonomous-driving, sampling-based-planning, lane-change

Trajectory planning in road-aligned coordinates, distance along a reference path and lateral offset from it, usually by sampling many smooth candidate trajectories and choosing the cheapest valid one.

Definition

Frenet frame planning describes a vehicle's motion relative to a reference curve, usually a lane centerline, instead of in global x and y. The position is given by the arc length s along the curve and the signed lateral offset d from it, measured in the moving tangent–normal frame of the curve, named after the Frenet–Serret formulas. In these coordinates, staying in lane means keeping d near zero, changing lanes means moving d to the next lane's offset, and following traffic means controlling s over time, which makes cost functions and constraints simple to write.

The widely used formulation is from Werling, Ziegler, Kammel and Thrun (ICRA 2010). It plans lateral and longitudinal motion separately as polynomials in time: quintic polynomials for lateral offset and for longitudinal tasks with a target position, such as following or stopping, and quartic polynomials for keeping a target speed. Each polynomial minimizes jerk between the current state and an end state. The planner samples many end states, such as different lateral offsets, arrival times and target speeds, combines lateral and longitudinal candidates, converts them back to Cartesian coordinates, and discards those that exceed curvature, acceleration or speed limits or collide with predicted obstacles. The valid candidate with the lowest cost, weighing jerk, time and deviation from the goal, is executed, and planning repeats at a high rate starting from the previous plan to keep successive trajectories consistent. At low speed the lateral offset is better planned as a function of s than of time.

The approach depends on a good reference line. The curve must be smooth with continuous curvature, and the conversion back to Cartesian coordinates breaks down when the lateral offset approaches the radius of curvature, which happens on tight bends with large offsets. Frenet frame planning suits structured roads and lanes; for parking lots or open spaces, planners such as Hybrid A are used instead.

Why it matters

Road driving is naturally described relative to lanes, and Frenet coordinates turn lane keeping, lane changes, merging and following into simple one-dimensional problems. The sampling formulation is easy to implement, gives smooth, comfortable trajectories, and has become a standard teaching and baseline planner for autonomous vehicles.

Key applications

  • Lane keeping, lane changes and overtaking on highways
  • Car following, merging and stopping at lines in urban driving
  • Baseline planners in autonomous-driving research platforms and simulators
  • Path following for robots moving along corridors, rows or tracks
  • Racing lines planned relative to a track centerline

In practice

On a highway, a self-driving test vehicle builds a smooth spline along the center of its current lane as the reference line. Each planning cycle, it samples lateral end offsets for its own lane and the neighboring lanes, several maneuver durations and a set of target speeds. It generates the matching polynomials, converts each candidate to x, y, heading and curvature, rejects those that exceed comfort or tire limits or come too close to the predicted paths of nearby cars, and passes the cheapest remaining trajectory to a tracking controller. A lane change happens when a candidate ending in the next lane becomes cheapest, for example when the car ahead slows down.

Common misconceptions

  • Frenet frame planning is a single algorithm — it is a coordinate choice; sampling with polynomials is the common approach, but optimization-based planners also work in Frenet coordinates
  • Trajectories are feasible because they are smooth in the Frenet frame — limits on curvature and acceleration must be checked after converting back to Cartesian coordinates
  • It works anywhere — it needs a smooth reference path and degrades on tight curves and in open spaces without lanes
  • Lateral and longitudinal planning are fully independent — they are planned separately for convenience but must be combined and checked together

Who should know this

  • Autonomous driving planning engineers
  • Researchers using driving simulators such as CARLA
  • Mobile robot developers following structured routes
  • Students learning trajectory generation

Related technologies

  • Quintic and quartic polynomial trajectories
  • Frenet–Serret formulas
  • Cubic spline reference lines
  • State lattice planner
  • Model predictive control
  • CARLA simulator

Related terms

  • Trajectory Planning
  • Bicycle model
  • Model Predictive Control
  • Obstacle Avoidance
  • Nonholonomic Constraint
  • Hybrid A*

Related projects

  • Ruckig: Real-time jerk-constrained time-optimal trajectory generation for robots and machines
  • Vision Pilot: Autoware Foundation's open-source L2 ADAS stack from a single camera
  • python_simple_mppi: MPPI control derived and implemented in numpy, with four worked examples
  • NVIDIA Alpamayo: NVIDIA's open driving VLA models that plan a path and explain it, with sim, RL and data

This website requires JavaScript to function properly. Please enable JavaScript to explore our collection of open source robotics projects.