Active Suspension Control Using LQR and Reinforcement Learning

A car suspension trades comfort against travel and road holding. This project compares a passive damper, a tuned damper, skyhook, LQR, LQG and TD3 reinforcement learning on one quarter-car model, one objective and one force limit, on generated ISO 8608 roads of classes A to E. This is a simulation: no vehicle, test rig or hardware test exists in this project. Road profiles come from an own simulation after ISO 8608 with fixed seeds; they are never measured data. Lowest J per class: class A: LQG (1.70, passive 1.98); class B: LQG (1.70, passive 1.96); class C: LQG (1.73, passive 1.99); class D: LQG (1.64, passive 1.94); class E: LQR (1.52, passive 2.03). Mean J of TD3 full against LQR: higher on classes A, B, C, D and E. Mean J of TD3 sensors against LQG: higher on classes A, B, C, D and E.

Category: Automobile Engineering

Automobile Engineering Project AU06 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.