My part of the course project for Intelligent Vehicles and Autonomous Driving. The agent controls a car's speed through intersections and traffic lights. Instead of following hand-written rules, it plans minimum-jerk motion primitives, which produce smooth, human-like acceleration profiles.
The controller is inspired by human sensorimotor behaviour:
- Action priming: generates candidate manoeuvres (affordances), following the minimal intervention principle.
- Action selection: picks the best candidate with a mechanism inspired by the basal ganglia.
Each manoeuvre is a closed-form minimum-jerk primitive: a quintic position profile defined by the initial state (v₀, a₀) and the target distance, speed and time.
| Scenario | Actions |
|---|---|
| Intersection | Stop at or before the stop line, or cross within a speed window while keeping a safe time gap to other vehicles |
| Traffic light | Stop at or before the light, or pass within a given time window and speed interval |
| Velocity / position profile | Primitive coefficients |
|---|---|
 |
 |
├── starting_point.cc # Agent main loop: receives scenario messages, sends manoeuvres
├── agent_functions.hpp # Stop / pass primitives, optimal time & velocity, low-level control
└── pass_primitive_plot.py # Plots the velocity and position profiles from logged coefficients
Key functions in agent_functions.hpp:
stopPrimitive: minimum-jerk stop at distancesfPassprimitive/Passprimitve_with_j0: pass within[Vmin, Vmax]and[Tmin, Tmax]choose_m_star: selects the primitive to executeLowLevelControl: converts the primitive into an acceleration command (with saturation)
Note: This repository contains only the parts I wrote. The simulator, server library and the other team members' code are not included, so it will not build on its own. Contact me if you would like more details.
C++ · Python · NumPy · Matplotlib · Optimal control · Motion primitives
No license is provided. This repository contains course material and code provided by the course instructors. Please contact me before reusing it.
Yogeswaran Amsavalli · GitHub