I am working on a project to implement a collision avoidance algorithm on a real UAV. I'm interested in understanding the process to set up a negative reward to account for scenarios wherein there is a UAV crash. This can be done very easily during simulation(if the UAV touches any object, the episode stops giving a negative reward). Any ideas will be highly appreciated.
From what I understand this is quite easy and you don't have to use reinforcement learning. You can also use standard ANNs.
For as long as you want:
If Touch (sensor. Maybe a button???) == 1
Update network (the network should output the probability of crashing in the next step. You can also use markov chains for multiple steps (just an idea))
I hope this answers your question.