game step

size of buffer

sample from buffer according to priority

send position command on local ENU

True if len(self.memory) >= self.batch_size

arg: - transition: the payload needed to be stored - prority: the priority of transition

determine if the uav is crashed. return: - True if it was crashed. - crash reward

size of buffer

send take-off command

update priorities.

send velocity command on body end

35 laser 1 vx 1 yaw_rate 1 ax 1 ayaw 1 distance 1 angle_dif

reset uav and target

reset target ball

add data to buffer

clear world

sample from buffer according to priority

check whether (x, y) falls into safe space. return: True - if not fall into

load data

load data

run recovery action (keep away from obstacles in slight velocity) return: - True if crash when recovering.

reset world

save data

save data

Embedding buffer is a container storing state data. The implementation here is for embedding layer constructing.

safe_space: safe_radius: wall_rate: thr propotion of having wall. cylidar_num:

initialize

prioritized replay buffer implemented by Pytorch

if no cmd, hold command is sent.

get iris state, and send vision.

determine whether action in state after time_step is valid return: - True if action is valid - valid reward