One of the fundamental challenges in reinforcement learning (RL) is the one of data efficiency: modern algorithms require a very large number of training samples, especially compared to humans, for solving environments with high-dimensional observations. The severity of this problem is increased when the reward signal is sparse. In this work, we propose learning a state representation in a self-supervised manner for reward prediction. The reward predictor learns to estimate either a raw or a smoothed version of the true reward signal in environment with a single, terminating, goal state. We augment the training of out-of-the-box RL agents by shaping the reward using our reward predictor during policy learning. Using our representation for preprocessing high-dimensional observations, as well as using the predictor for reward shaping, is shown to significantly enhance Actor Critic using Kronecker-factored Trust Region and Proximal Policy Optimization in single-goal environments with visual inputs.

Bolstering multi-agent learning algorithms to tackle complex coordination and control tasks has been a long-standing challenge of on-going research. Numerous methods have been proposed to help reduce the effects of non-stationarity and unscalability. In this work, we investigate a novel approach to decentralized multi-agent learning and planning that attempts to address these two challenges. In particular, this method is inspired by the cohesion, coordination, and behavior of ant colonies. As a result, these algorithms are designed to be naturally scalable to systems with numerous agents. While no optimality is guaranteed, the method is intended to work well in practice and scale better in efficacy with the number of agents present than others. The approach combines single-agent RL and an ant-colony-inspired decentralized, stigmergic algorithm for multi-agent path planning and environment modification. Specifically, we apply this algorithm in a setting where agents must navigate to a goal location, learning to push rectangular boxes into holes to yield new traversable pathways. It is shown that while the approach yields promising success in this particular environment, it may not be as easily generalized to others. The algorithm designed is notably scalable to numerous agents but is limited in its performance due to its relatively simplistic, rule-based approach. Furthermore, the composability of RL-trained policies is called into question, where, while policies are successful in their training environments, applying trained policies to a larger-scale, multi-agent framework results in unpredictable behavior.

In this project, we designed an intelligent assistant player for the single-player game Space Invaders with the aim to provide a satisfying co-op experience. The agent behaviour was designed using reinforcement learning techniques and evaluated based on several criteria. We validate the hypothesis that an AI-driven computer player can provide a satisfying co-op experience.

A carefully engineered radiofrequency (RF) pulse plays a key role in a number of systems such as mobile phone, radar, and magnetic resonance imaging (MRI). The design of an RF waveform, however, is often posed as an inverse problem that has no general solution. As a result, various design methods each with a specific purpose have been developed based on the intuition of human experts. In this work, we propose an artificial intelligence-powered RF pulse design framework, DeepRF, which utilizes the self-learning characteristics of deep reinforcement learning (DRL) to generate a novel RF beyond human intuition. Additionally, the method can design various types of RF pulses via customized reward functions. The algorithm of DeepRF consists of two modules: the RF generation module, which utilizes DRL to explore new RF pulses, and the RF refinement module, which optimizes the seed RF pulses from the generation module via gradient ascent. The effectiveness of DeepRF is demonstrated using four exemplary RF pulses, slice-selective excitation pulse, slice-selective inversion pulse, B1-insensitive volume inversion pulse, and B1-insensitive selective inversion pulse, that are commonly used in MRI. The results show that the DeepRF-designed pulses successfully satisfy the design criteria while improving specific absorption rates when compared to those of the conventional RF pulses. Further analyses suggest that the DeepRF-designed pulses utilize new mechanisms of magnetization manipulation that are difficult to be explained by conventional theory, suggesting the potentials of DeepRF in discovering unseen design dimensions beyond human intuition. This work may lay the foundation for an emerging field of AI-driven RF waveform design.

Deep reinforcement learning methods have shown great performance on many challenging cooperative multi-agent tasks. Two main promising research directions are multi-agent value function decomposition and multi-agent policy gradients. In this paper, we propose a new decomposed multi-agent soft actor-critic (mSAC) method, which effectively combines the advantages of the aforementioned two methods. The main modules include decomposed Q network architecture, discrete probabilistic policy and counterfactual advantage function (optinal). Theoretically, mSAC supports efficient off-policy learning and addresses credit assignment problem partially in both discrete and continuous action spaces. Tested on StarCraft II micromanagement cooperative multiagent benchmark, we empirically investigate the performance of mSAC against its variants and analyze the effects of the different components. Experimental results demonstrate that mSAC significantly outperforms policy-based approach COMA, and achieves competitive results with SOTA value-based approach Qmix on most tasks in terms of asymptotic perfomance metric. In addition, mSAC achieves pretty good results on large action space tasks, such as 2c vs 64zg and MMM2.

In many deep reinforcement learning settings, when an agent takes an action, it repeats the same action a predefined number of times without observing the states until the next action-decision point. This technique of action repetition has several merits in training the agent, but the data between action-decision points (i.e., intermediate frames) are, in effect, discarded. Since the amount of training data is inversely proportional to the interval of action repeats, they can have a negative impact on the sample efficiency of training. In this paper, we propose a simple but effective approach to alleviate to this problem by introducing the concept of pseudo-actions. The key idea of our method is making the transition between action-decision points usable as training data by considering pseudo-actions. Pseudo-actions for continuous control tasks are obtained as the average of the action sequence straddling an action-decision point. For discrete control tasks, pseudo-actions are computed from learned action embeddings. This method can be combined with any model-free reinforcement learning algorithm that involves the learning of Q-functions. We demonstrate the effectiveness of our approach on both continuous and discrete control tasks in OpenAI Gym.

It has recently been shown that reinforcement learning can be used to train generators capable of producing high-quality game levels, with quality defined in terms of some user-specified heuristic. To ensure that these generators' output is sufficiently diverse (that is, not amounting to the reproduction of a single optimal level configuration), the generation process is constrained such that the initial seed results in some variance in the generator's output. However, this results in a loss of control over the generated content for the human user. We propose to train generators capable of producing controllably diverse output, by making them "goal-aware." To this end, we add conditional inputs representing how close a generator is to some heuristic, and also modify the reward mechanism to incorporate that value. Testing on multiple domains, we show that the resulting level generators are capable of exploring the space of possible levels in a targeted, controllable manner, producing levels of comparable quality as their goal-unaware counterparts, that are diverse along designer-specified dimensions.

Deep reinforcement learning (DRL) has recently shown its success in tackling complex combinatorial optimization problems. When these problems are extended to multiobjective ones, it becomes difficult for the existing DRL approaches to flexibly and efficiently deal with multiple subproblems determined by weight decomposition of objectives. This paper proposes a concise meta-learning-based DRL approach. It first trains a meta-model by meta-learning. The meta-model is fine-tuned with a few update steps to derive submodels for the corresponding subproblems. The Pareto front is built accordingly. The computational experiments on multiobjective traveling salesman problems demonstrate the superiority of our method over most of learning-based and iteration-based approaches.

It is commonly claimed that AI will replace most manual labor in the future; however, is this really the case? AI technologies do have higher image recognition accuracy compared to humans in some limited contexts, and have consistently outperformed humans in classical games such as Go and chess. Nonetheless, we believe that even advanced future developments based on current technology will not lead to robots replacing humans. AI systems' fundamental lack of ability to communicate naturally and effectively with humans is among the most significant reasons that they cannot replace human labor. Here, one may believe that such communication could be achieved via the development of natural language processing (NLP) technology [4]; however, NLP technologies are systems for estimating the content of human statements and their meanings; they do not constitute communication. That is, humans do not feel that robots using such systems truly understand and respond to them appropriately. Therefore, if effective communication is not achieved, robots will continue to function only as tools to assist humans. Advancements improving the accuracy or effectiveness of various specific tasks do not indicate that robots are equivalent to human beings. Under this scenario, how can we enable robots to communicate with humans?

The separation assurance task will be extremely challenging for air traffic controllers in a complex and high density airspace environment. Deep reinforcement learning (DRL) was used to develop an autonomous separation assurance framework in our previous work where the learned model advised speed maneuvers. In order to improve the safety of this model in unseen environments with uncertainties, in this work we propose a safety module for DRL in autonomous separation assurance applications. The proposed module directly addresses both model uncertainty and state uncertainty to improve safety. Our safety module consists of two sub-modules: (1) the state safety sub-module is based on the execution-time data augmentation method to introduce state disturbances in the model input state; (2) the model safety sub-module is a Monte-Carlo dropout extension that learns the posterior distribution of the DRL model policy. We demonstrate the effectiveness of the two sub-modules in an open-source air traffic simulator with challenging environment settings. Through extensive numerical experiments, our results show that the proposed sub-safety modules help the DRL agent significantly improve its safety performance in an autonomous separation assurance task.