Adaptive Cruise Control Using MPC and Reinforcement Learning

Adaptive cruise control (ACC) keeps a safe time gap to the car in front while respecting comfort limits on acceleration and jerk. This project simulates the longitudinal dynamics of a passenger car with a first-order powertrain lag, drives the car in front with the real speed traces of six official US EPA drive cycles, and compares a PD controller (the specification's gains and gains tuned on the training cycles), a linear model-predictive controller (MPC) with explicit acceleration limits and a soft minimum gap, and TD3 and PPO reinforcement-learning agents - split by drive cycle, so that the test cycles were never seen during tuning or training. Robustness tests (a slower or quicker powertrain, sensor noise, cars cutting in) and a time-gap sweep (safety against lane capacity) complete the study. A web application simulates any controller on any cycle or uploaded lead-speed log, compares controllers, manages vehicle variants, time-gap settings with approval and safety events, on five databases. The MPC is cross-checked with CasADi and do-mpc, the MATLAB code with GNU Octave, and a complete MATLAB / Simulink version is included.

Category: Automobile Engineering

Automobile Engineering Project AU01 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.