Inverse reinforcement learning for autonomous robot control
The ability of a robot to autonomously adapt to an environment plays an important role in the coexistence of humans and robots. In recent years, inverse reinforcement learning, which estimates a reward function to be used for reinforcement learning from demonstrations of experts, has attracted attention in the field of robot control. For the purpose of reducing the burden on experts, we propose an inverse reinforcement learning method that estimates a reward function from the scores given by an expert for arbitrary trajectories of a robot.