April 2018
Intermediate to advanced
334 pages
10h 18m
English
Reward function is of key importance in specifying the objective of the learning agent in robot reinforcement learning. As we have learned, for reinforcement learning algorithms, the ultimate objective is to maximize the expected sum of rewards from the start state till the goal state is reached.
In a real-world scenario, devising a good reward function is a big challenge. Therefore, representing or specifying a goal is a challenge in real-world scenarios. The real-world environment is full of uncertainty therefore, the reward function should be able to capture the positive state associated with such uncertainty.
Some domains receive rewards after task completion, where uncertainty is ...