Learning based model predictive Controller (LB MPC) for Altitude Control of VTOL aircraft
Proposed a DDPG based LBMPC controller to learn the weights in a mpc cost formulation
Bachelors final year project
A detailed description can be found in my bachelor’s final year project pdf (chapter4)
pdf link:
Bachelor_thesis
Naive ddpg based lbmpc controller
avg reward per episode vs iteration
Aircraft was able to reach 103m ,-2m less than desired height
After 1000 iteration we can see increase in the avg reward per episode from 0.48 to 0.62 , and thus algorithm was able to learn some better weights for mpc . We can see in below fig the aircraft can now reach to 104.5m ,1m less than the desired height
Aircraft was able to reach 104.5m ,0.5m less than desired height