Open Problems and Modern Solutions for Deep Reinforcement Learning
This work addresses incremental improvements in deep reinforcement learning for researchers and practitioners dealing with data inefficiency and reward design issues.
The paper reviews two solutions to address data inefficiency and inflexible reward design in deep reinforcement learning: one combines extrinsic and intrinsic rewards for human-robot collaboration to improve task performance and obstacle avoidance, and the other uses selective attention and particle filters to enhance efficiency and flexibility by attending to pre-learned features.
Deep Reinforcement Learning (DRL) has achieved great success in solving complicated decision-making problems. Despite the successes, DRL is frequently criticized for many reasons, e.g., data inefficient, inflexible and intractable reward design. In this paper, we review two publications that investigate the mentioned issues of DRL and propose effective solutions. One designs the reward for human-robot collaboration by combining the manually designed extrinsic reward with a parameterized intrinsic reward function via the deterministic policy gradient, which improves the task performance and guarantees a stronger obstacle avoidance. The other one applies selective attention and particle filters to rapidly and flexibly attend to and select crucial pre-learned features for DRL using approximate inference instead of backpropagation, thereby improving the efficiency and flexibility of DRL. Potential avenues for future work in both domains are discussed in this paper.