Reinforcement learning strategies for severely unbalanced drone
Quadcopter drones rely entirely on their four rotors to control altitude and attitude. A complete failure of any single rotor results in losing stability and control unless the controller can reconfigure the drone’s remaining actuators to re-establish balance in forces and moments. Previous attempts...
| Main Author: | |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | English |
| Published: |
IEEE
2024
|
| Online Access: | http://psasir.upm.edu.my/id/eprint/118419/ http://psasir.upm.edu.my/id/eprint/118419/1/118419.pdf |
| Summary: | Quadcopter drones rely entirely on their four rotors to control altitude and attitude. A complete failure of any single rotor results in losing stability and control unless the controller can reconfigure the drone’s remaining actuators to re-establish balance in forces and moments. Previous attempts to reinstate full stability and control for this type of unbalanced drone by using classical and modern control laws have been unsuccessful. Recent development in Artificial Intelligence technology, however, may suggest a new control solution to rescue the crippled drone. The work reported in this paper proposes the initial strategies to implement Reinforcement Learning technique to find an optimal solution by regulating the remaining rotor speeds. The mode chosen for the study is hovering. The strategy to train the Reinforcement Learning agent is also included. |
|---|