One obstacle to applying reinforcement learning algorithms to real-world problems is the lack of suitable reward functions. Designing such reward functions is difficult in part because the user only has an implicit understanding of the task objective. This gives rise to the agent alignment problem: how do we create agents that behave in accordance with the user's intentions? We outline a high-level research direction to solve the agent alignment problem centered around reward modeling: learning a reward function from interaction with the user and optimizing the learned reward function with reinforcement learning. We discuss the key challenges we expect to face when scaling reward modeling to complex and general domains, concrete approaches to mitigate these challenges, and ways to establish trust in the resulting agents.

We present an experimental realization of a measurement-based adaptation protocol with quantum reinforcement learning in a Rigetti cloud quantum computer. The experiment in this few-qubit superconducting chip faithfully reproduces the theoretical proposal, setting the first steps towards a semiautonomous quantum agent. This experiment paves the way towards quantum reinforcement learning with superconducting circuits.

Training task-completion dialogue agents with reinforcement learning usually requires a large number of real user experiences. The Dyna-Q algorithm extends Q-learning by integrating a world model, and thus can effectively boost training efficiency using simulated experiences generated by the world model. The effectiveness of Dyna-Q, however, depends on the quality of the world model - or implicitly, the pre-specified ratio of real vs. simulated experiences used for Q-learning. To this end, we extend the recently proposed Deep Dyna-Q (DDQ) framework by integrating a switcher that automatically determines whether to use a real or simulated experience for Q-learning. Furthermore, we explore the use of active learning for improving sample efficiency, by encouraging the world model to generate simulated experiences in the state-action space where the agent has not (fully) explored. Our results show that by combining switcher and active learning, the new framework named as Switch-based Active Deep Dyna-Q (Switch-DDQ), leads to significant improvement over DDQ and Q-learning baselines in both simulation and human evaluations.

Well structured visual representations can make robot learning faster and can improve generalization. In this paper, we study how we can acquire effective object-centric representations for robotic manipulation tasks without human labeling by using autonomous robot interaction with the environment. Such representation learning methods can benefit from continuous refinement of the representation as the robot collects more experience, allowing them to scale effectively without human intervention. Our representation learning approach is based on object persistence: when a robot removes an object from a scene, the representation of that scene should change according to the features of the object that was removed. We formulate an arithmetic relationship between feature vectors from this observation, and use it to learn a representation of scenes and objects that can then be used to identify object instances, localize them in the scene, and perform goal-directed grasping tasks where the robot must retrieve commanded objects from a bin. The same grasping procedure can also be used to automatically collect training data for our method, by recording images of scenes, grasping and removing an object, and recording the outcome. Our experiments demonstrate that this self-supervised approach for tasked grasping substantially outperforms direct reinforcement learning from images and prior representation learning methods.

Model-free reinforcement learning methods such as the Proximal Policy Optimization algorithm (PPO) have successfully applied in complex decision-making problems such as Atari games. However, these methods suffer from high variances and high sample complexity. On the other hand, model-based reinforcement learning methods that learn the transition dynamics are more sample efficient, but they often suffer from the bias of the transition estimation. How to make use of both model-based and model-free learning is a central problem in reinforcement learning. In this paper, we present a new technique to address the trade-off between exploration and exploitation, which regards the difference between model-free and model-based estimations as a measure of exploration value. We apply this new technique to the PPO algorithm and arrive at a new policy optimization method, named Policy Optimization with Model-based Explorations (POME). POME uses two components to predict the actions' target values: a model-free one estimated by Monte-Carlo sampling and a model-based one which learns a transition model and predicts the value of the next state. POME adds the error of these two target estimations as the additional exploration value for each state-action pair, i.e, encourages the algorithm to explore the states with larger target errors which are hard to estimate. We compare POME with PPO on Atari 2600 games, and it shows that POME outperforms PPO on 33 games out of 49 games.

The idea of reusing or transferring information from previously learned tasks (source tasks) for the learning of new tasks (target tasks) has the potential to significantly improve the sample efficiency of a reinforcement learning agent. In this work, we describe a novel approach for reusing previously acquired knowledge by using it to guide the exploration of an agent while it learns new tasks. In order to do so, we employ a variant of the growing self-organizing map algorithm, which is trained using a measure of similarity that is defined directly in the space of the vectorized representations of the value functions. In addition to enabling transfer across tasks, the resulting map is simultaneously used to enable the efficient storage of previously acquired task knowledge in an adaptive and scalable manner. We empirically validate our approach in a simulated navigation environment, and also demonstrate its utility through simple experiments using a mobile micro-robotics platform. In addition, we demonstrate the scalability of this approach, and analytically examine its relation to the proposed network growth mechanism. Further, we briefly discuss some of the possible improvements and extensions to this approach, as well as its relevance to real world scenarios in the context of continual learning.

The class of Gaussian Process (GP) methods for Temporal Difference learning has shown promise for data-efficient model-free Reinforcement Learning. In this paper, we consider a recent variant of the GP-SARSA algorithm, called Sparse Pseudo-input Gaussian Process SARSA (SPGP-SARSA), and derive recursive formulas for its predictive moments. This extension promotes greater memory efficiency, since previous computations can be reused and, interestingly, it provides a technique for updating value estimates on a multiple timescales

Multi-agent learning provides a potential framework for learning and simulating traffic behaviors. This paper proposes a novel architecture to learn multiple driving behaviors in a traffic scenario. The proposed architecture can learn multiple behaviors independently as well as simultaneously. We take advantage of the homogeneity of agents and learn in a parameter sharing paradigm. To further speed up the training process asynchronous updates are employed into the architecture. While learning different behaviors simultaneously, the given framework was also able to learn cooperation between the agents, without any explicit communication. We applied this framework to learn two important behaviors in driving: 1) Lane-Keeping and 2) Over-Taking. Results indicate faster convergence and learning of a more generic behavior, that is scalable to any number of agents. When compared the results with existing approaches, our results indicate equal and even better performance in some cases.

Discounted-sum games provide a formal model for the study of reinforcement learning, where the agent is enticed to get rewards early since later rewards are discounted. When the agent interacts with the environment, she may regret her actions, realizing that a previous choice was suboptimal given the behavior of the environment. The main contribution of this paper is a PSPACE algorithm for computing the minimum possible regret of a given game. To this end, several results of independent interest are shown. (1) We identify a class of regret-minimizing and admissible strategies that first assume that the environment is collaborating, then assume it is adversarial---the precise timing of the switch is key here. (2) Disregarding the computational cost of numerical analysis, we provide an NP algorithm that checks that the regret entailed by a given time-switching strategy exceeds a given value. (3) We show that determining whether a strategy minimizes regret is decidable in PSPACE.

We want to make progress toward artificial general intelligence, namely general-purpose agents that autonomously learn how to competently act in complex environments. The purpose of this report is to sketch a research outline, share some of the most important open issues we are facing, and stimulate further discussion in the community. The content is based on some of our discussions during a weeklong workshop held in Barbados in February 2018. We adopt the reinforcement learning (RL) formulation, where an agent interacts sequentially with an environment, and the agent is provided a reward signal that unambiguously defines success. We want to explicitly consider some of the most challenging dimensions for a developing intelligence.