We present a hierarchical reinforcement learning framework that formulates each task in the hierarchy as a special type of Markov decision process for which the Bellman equation is linear and has analytical solution. Problems of this type, called linearly-solvable MDPs (LMDPs) have interesting properties that can be exploited in a hierarchical setting, such as efficient learning of the optimal value function or task compositionality. The proposed hierarchical approach can also be seen as a novel alternative to solving LMDPs with large state spaces. We derive a hierarchical version of the so-called Z-learning algorithm that learns different tasks simultaneously and show empirically that it significantly outperforms the state-of-the-art learning methods in two classical HRL domains: the taxi domain and an autonomous guided vehicle task.

Taxis (which include cars working with car aggregation systems such as Uber, Grab, Lyft etc.) have become a critical component in the urban transportation. While most research and applications in the context of taxis have focused on improving performance from a customer perspective, in this paper, we focus on improving performance from a taxi driver perspective. Higher revenues for taxi drivers can help bring more drivers into the system thereby improving availability for customers in dense urban cities. Typically, when there is no customer on board, taxi drivers will cruise around to find customers either directly (on the street) or indirectly (due to a request from a nearby customer on phone or on aggregation systems). For such cruising taxis, we develop a Reinforcement Learning (RL) based system to learn from real trajectory logs of drivers to advise them on the right locations to find customers which maximize their revenue. There are multiple translational challenges involved in building this RL system based on real data, such as annotating the activities (e.g., roaming, going to a taxi stand, etc.) observed in trajectory logs, identifying the right features for a state, action space and evaluating against real driver performance observed in the dataset. We also provide a dynamic abstraction mechanism to improve the basic learning mechanism. Finally, we provide a thorough evaluation on a real world data set from a developed Asian city and demonstrate that an RL based system can provide significant benefits to the drivers.

Video games have proved to be a very defying laboratory to study machine-learning techniques, such as Deep Reinforcement Learning (DRL) algorithms. This paper presents a new approach for a DRL-based agent trained through Deep Q-Network (DQN) technique to perform free kicks in FIFA 18 game. The main motivation for choosing this case study is the fact that, like in many situations of the real life, FIFA represents a stochastic environment. Coping with this task, the main contributions of the present paper consist on: inspired on the OpenAI Gym and on the OpenAI Universe platforms, implementing a new user-friendly interface (in terms of portability and use simplicity) to connect the learning module with the 3D FIFA's game environment; implementing a DRL-based agent for free kicks in FIFA that uses two distinct data representations retrieved from lower cost computational procedures. The results were validated through two evaluative parameters: score of well succeed kicks and training time.

Learning strategies to address problems on graph and tree structures with no a-priori size limitations in cases where no known solution exists (and thus supervised data is hard to obtain), is a difficult problem with potential applications in a wide range of domains ranging from biological networks to protein folding and social network search. The main challenges here arise from the variable size representation that needs to be resolved in the context of Reinforcement Learning (RL) to address the problem. In this paper we consider a common, specific tree problem and show that it can be addressed using a combination of feature engineering and carefully designed learning processes. In particular, We consider the classical Nearest Neighbor Interchange (NNI) distance between unrooted labeled trees, which is defined as the minimum-cost sequence of operations that transform one tree into another. We introduce a representation and a reinforcement learning method that learns the transition dynamics and iteratively changes an arbitrary initial labeled tree into a goal configuration reachable through NNI. The differential tree representation and NNI actions permits the system to learn a strategy that is applicable to arbitrary sized trees. To train the system, we introduce a training process that uses randomly sampled trajectories to incrementally train more and more complex problems to overcome the difficulty of the overall strategy space. Experiments performed show that the system can successfully learn a strategy for effective NNI on complex trees.

During opinion formation, interacting agents can be assumed to be engaging in learning and decision-making processes to satisfy their individual goals. These goals are determined by the agents' preferences – which are often unknown, complex and unpredictable. Most opinion formation frameworks however, assume static preferences and fail to model practical situations where human preferences change. We propose a new framework to simulate the process of opinion formation under uncertainty and dynamism. Agents who are unaware of their implicit con-textual preferences utilize inverse reinforcement learning to compute reward functions that determines their preferences. Reinforcement learning is subsequently used to optimize the agents' behavior and satisfy their individual goals. The novelty of our approach lies in its ability to capture uncertainty and dynamism in the agent's preferences, which are assumed to be unknown initially. This framework is compared to a baseline method based on reinforcement learning, and results show its ability to per-form better under dynamic scenarios.

Recent progress in AI and Reinforcement learning has shown great success in solving complex problems with high dimensional state spaces. However, most of these successes have been primarily in simulated environments where failure is of little or no consequence. Most real-world applications, however, require training solutions that are safe to operate as catastrophic failures are inadmissible especially when there is human interaction involved. Currently, Safe RL systems use human oversight during training and exploration in order to make sure the RL agent does not go into a catastrophic state. These methods require a large amount of human labor and it is very difficult to scale up. We present a hybrid method for reducing the human intervention time by combining model-based approaches and training a supervised learner to to improve sample efficiency while also ensuring safety. We evaluate these methods on various grid-world environments using both standard and visual representations and show that our approach achieves better performance in terms of sample efficiency, number of catastrophic states reached as well as overall task performance compared to traditional model-free approaches.

Reinforcement learning (RL) agents use trial and error to learn action policies for environment states. Environments with continuous action spaces are far more challenging for RL than those with discrete actions because there are infinite possible continuous action values from which to choose. Dynamic environments create additional challenges for RL agents, which must adjust rapidly to changes. We recently introduced REINFORCE SUN, a superclass of REINFORCE with Gaussian units, that allows for stochasticity at different levels of granularity in artificial neural networks (synapse, unit, or network), and have shown that moving stochasticity to synapses greatly aids RL in both static and dynamic environments with continuous action spaces. However, we also found that performance in dynamic environments remained substantially lower than desired.

Reinforcement Learning has the limitation that problems become too large very quickly. Dividing the problem into a hierarchy of subtasks allows for a strategy of divide and conquer, which is what makes Hierarchical Reinforcement Learning (HRL) algorithms often more efficient at finding solutions quicker than more naive approaches. One of the biggest challenges with HRL is the construction of a hierarchy to be used by the algorithm. Hierarchies are often designed by a person using their own knowledge of the problem. We propose method for automatically discovering task hierarchies based on a data mining technique, Association Rule Learning (ARL).

In this paper, we present a novel model-free reinforcement learning approach for solving a conventional leader-follower problem using autonomous wheeled mobile robots. Specifically, the proposed learning approach will determine the linear velocity and the steering angle (control actions) of a follower robot so that it can follow the time-varying motion trajectory of a leader robot. The setup of the online adaptive learning mechanism does not rely on any dynamical or kinematic parameters, i.e., model-free'', of the considered car-like robots. Bellman's principle of optimality is employed to approximate the reward of the control actions determined by the proposed model-free adaptive learning algorithm. A set of computer experiments has been conducted to evaluate the performance of the proposed algorithm under various unplanned leader-trajectories.

Weakly-coupled Markov decision processes can be decomposed into subprocesses that interact only through a small set of bottleneck states. We study a hierarchical reinforcement learning algorithm designed to take advantage of this particular type of decomposability. To test our algorithm, we use a decision-making problem faced by autonomous planetary rovers. In this problem, a Mars rover must decide which activities to perform and when to traverse between science sites in order to make the best use of its limited resources. In our experiments, the hierarchical algorithm performs better than Q-learning in the early stages of learning, but unlike Q-learning it converges to a suboptimal policy. This suggests that it may be advantageous to use the hierarchical algorithm when training time is limited.