33th Congress of the International Council of the Aeronautical Sciences

06.2 - Flight Dynamics and Control (UAV related)

SAFE REINFORCEMENT LEARNING BASED MULTI-ROTOR COLLISION AVOIDANCE WITH UNEXPECTED OBSTACLES

H.J. Ahn╣, C.S. Lim╣, D.W. Lee╣, H.C. Bang╣; ╣Korea Advanced Institute of Science and Technology(KAIST), South Korea

This paper suggest the safe RL for a quadrotor collision avoidance. The main contribution of this paper are 1) introducing the convex safety filter to guarantee the non-collision flight with RL framework, and 2) using simplified depth image to make easier application of the RL framework generated through the simulation to the real world, and to reduce the calculation time.


View Paper