Search

One of the main concerns of deep reinforcement learning (DRL) is the data inefficiency problem, which stems both from an inability to fully utilize data acquired and from naive exploration strategies. In order to alleviate these problems, we propose a DRL algorithm that aims to improve data efficiency via both the utilization of unrewarded experiences and the exploration strategy by combining ideas from unsupervised auxiliary tasks, intrinsic motivation, and hierarchical reinforcement learning (HRL). Our method is based on a simple HRL architecture with a metacontroller and a subcontroller. The subcontroller is intrinsically motivated by the metacontroller to learn to control aspects of the environment, with the intention of giving the agent: 1) a neural representation that is generically useful for tasks that involve manipulation of the environment and 2) the ability to explore the environment in a temporally extended manner through the control of the metacontroller. In this way, we reinterpret the notion of pixel- and feature-control auxiliary tasks as reusable skills that can be learned via an intrinsic reward. We evaluate our method on a number of Atari 2600 games. We found that it outperforms the baseline in several environments and significantly improves performance in one of the hardest games—Montezuma’s revenge—for which the ability to utilize sparse data is key. We found that the inclusion of intrinsic reward is crucial for the improvement in the performance and that most of the benefit seems to be derived from the representations learned during training.

Many CPG-based locomotion models have a problem known as the tracking error problem, where the mismatch between the CPG driving signal and the state of the robot can cause undesirable behaviours for legged robots. Towards alleviating this problem, we introduce a mechanism that modulates the CPG signal using the robot’s interoceptive information. The key concept is to generate a driving signal that is easier for the robot to follow, yet can drive the locomotion of the robot. This can be done by nudging the CPG signal in the direction of lower tracking error, which can be analytically calculated. Unlike other reactive CPG, the proposed method does not rely on any parametric learning ability to adjust the shape of the signal, making it a unique option for a biological adaptive motor control. Our experiment results show that the proposed method successfully reduces the tracking error. We also show that the CPG signal, regulated by the proposed method, is robust to perturbation and can smoothly return back to the default pattern.

Overfitting to learning partners is a known problem, in multi-agent reinforcement learning (MARL), due to the co-evolution of learning agents. Previous works explicitly add diversity to learning partners for mitigating this problem. However, since there are many approaches for introducing diversity, it is not clear which one should be used under what circumstances. In this work, we clarify the situation and reveal that widely used methods such as partner sampling and population-based training are unreliable at introducing diversity under fully cooperative multi-agent Markov decision process. We find that generating pre-trained partners is a simple yet effective procedure to achieve diversity. Finally, we highlight the impact of diversified learning partners on the generalization of learning agents using cross-play and ad-hoc team performance as evaluation metrics.