search

RoblabWh / RobLearn

ROS Robotic Deep Learning
171 stars 59 forks source link

reward function design #4

Open Joll123 opened 5 years ago

Joll123 commented 5 years ago

Does your reward function contain obstacle distance information?

ElhadjHoussem commented 5 years ago

Hi, The distance to obstacles is what the Neuronal network actually learning from as Input , which are the Laser signal ( Normalized [0-1]) beside the one-hot vector of oriontation. It could be added to the reward, but i think it will disturb the information that lead to the goal, and it will make the robot more rotating around himself, cause it can bring him more hight reward from that.

Joll123 commented 5 years ago

thank you for your reply. I am working on obstacle avoidance mobile robot navigation. I added obstacle distance information to the reward, making it difficult for the neural network to find the target.